Cell-type specific transcriptional networks in root xylem adjacent cell layers

Transport of water, ions and signals from roots to leaves via the xylem vessels is essential for plant life and needs to be tightly regulated. The final composition of the transpiration stream before passage into the shoots is controlled by the xylem-adjacent cell layers, namely xylem parenchyma and pericycle, in the upper part of the root. To unravel regulatory networks in this strategically important location, we generated Arabidopsis lines expressing a nuclear tag under the control of the HKT1 promoter. HKT1 retrieves sodium from the xylem to prevent toxic levels in the shoot, and this function depends on its specific expression in upper root xylem-adjacent tissues. Based on FACS RNA-sequencing and INTACT ChIP-sequencing, we identified the gene repertoire that is preferentially expressed in the tagged cell types and discovered transcription factors experiencing cell-type specific loss of H3K27me3 demethylation. For one of these, ZAT6, we show that H3K27me3-demethylase REF6 is required for de-repression. Analysis of zat6 mutants revealed that ZAT6 activates a suite of cell-type specific downstream genes and restricts Na+ accumulation in the shoots. The combined Files open novel opportunities for ‘bottom-up’ causal dissection of cell-type specific regulatory networks that control root-to-shoot communication under environmental challenge.


INTRODUCTION 1
Multicellular organisms are composed of different organs, tissues and cell types, which carry out 2 distinct functions. In plants, the leaves are responsible for the assimilation of carbon while the roots 3 take up water and mineral nutrients (Amtmann and Blatt, 2009). Within the roots, there is a further 4 division of tasks along both the radial and the longitudinal axis ( Barberon and Geldner, 2014). 5 Transport of water and minerals occurs first in a radial manner from the epidermis towards the 6 central cylinder (stele). Uptake into the xylem vessels then enables upwards longitudinal transport 7 with the transpiration stream (Lucas et al., 2013). In the mature parts of the root system, the 8 movement of substances into and out of the stele involves at least one trans-membrane step into the 9 symplast because the apoplastic pathway is blocked by the Casparian strip and suberin barriers 10 ( Barberon and Geldner, 2014). A second trans-membrane step is required for loading and unloading 11 of the (apoplastic) xylem vessels (Møller et al., 2009). It occurs either at the pericycle (Casimiro et 12 al., 2003) or in the cells that are directly adjacent to the xylem, often called the xylem parenchyma 13 (Maathuis et al., 2014) (Møller et al., 2009. Control of passage in these tissues is essential for 14 regulating the content of the transpiration stream, which moves not only water and nutrients to the 15 shoot, but also toxic substances (Mendoza-Cózatl et al., 2011;Munns and Tester, 2008) and signals 16 (Notaguchi and Okamoto, 2015). 17

Specific gene expression in the root stele underpins important functions. For example, in 18
Arabidopsis thaliana, several membrane transporters have been identified that either release or 19 retrieve K + or Na + into and from the transpiration stream, including the K + channel SKOR1 20 (Sharma et al., 2013), K + /H + antiporter NPF7.3 (Li et al., 2017) and the Na + transporters SOS1 (Shi 21 et al., 2002) and HKT1 (Munns and Tester, 2008;Møller et al., 2009). All of them show 22 preferential expression in root xylem-adjacent cells. Regulation of xylem content also relies on cell-23 type specific signals and signal perception. For example, induction of a NADPH oxidase in the root 24 xylem parenchyma under salt stress leads to local generation of reactive oxygen species (ROS) and 25 reduced root-to-shoot Na + delivery (Jiang et al., 2013). The xylem Na + /K + ratio is regulated by a 26 radial ethylene signal from the root periphery to the stele, which indicates cell-type specific signal 27 perception in the xylem-adjacent cell layers (Jiang et al., 2013). However, the individual 28 components and targets of the regulatory network in this strategically important location remain to 29 be elucidated. 30 To better understand the processes that control root-shoot communication we need to identify 31 the genes expressed in the relevant cell types and we need to investigate their causal inter-32 relationships. FACS combined with microarray analysis or RNA-Sequencing (Birnbaum et al., 33 2003;Brady et al., 2007;Dinneny et al., 2008;Walker et al., 2017)  primary mechanism since they are important for transcriptional regulation during development 12 (Roudier et al., 2011;Liu et al., 2010). H3K27me3 is a hallmark of gene repression (Zhang et al., 13 2007) and different H3K27me3 marking has been reported for individual plant cell types, such as 14 root epidermal hair and non-hair cells (Deal and Henikoff, 2010), vascular cells (de Lucas et al.,15 2016) and guard cells (Lee et al., 2019). Moreover, several studies point to a role of H3K27me3 in 16 establishing and maintaining cellular identity (Ikeuchi et al., 2015). However, recent study on guard 17 cell lineages highlighted that the number of differentially H3K27me3-marked genes was small 18 compared to the number of differentially expressed genes, and the authors proposed that epigenetic 19 re-programming of a few core regulators could proliferate cell-type specificity into a large cell-type 20 specific transcriptome (Lee et al. (2019). 21 Here we used a combination of FACS-RNA-seq and INTACT-ChIP-seq on Arabidopsis 22 thaliana root samples to address the following questions: 1. Which genes are preferentially 23 expressed in cell-types that control root xylem content? 2. What determines their cell-type specific 24 expression? 3. Which core regulators control the cell-type specific transcriptional network? 25 Our results show that H3K27me3-mediated de-repression of a small set of transcription factors 26 is sufficient for establishing large cell-type specific transcriptional networks, and we identify ZAT6 27 as one histone demethylation-dependent regulatory hub in root xylem-adjacent cell types. 28

RESULTS 30
Cell-tagging driven by the HKT1 promoter identifies co-expressed genes 31 For sorting of cell types controlling xylem content in the roots we took advantage of the cell-type 32 specific expression and function of HKT1 (At4g10310), a plasma membrane transport protein that 33 retrieves Na + from the transpiration stream (Sunarpi et al., 2005, Møller et al., 2009. To enable 34 analysis of the HKT1-expressing cell types by FACS and INTACT we generated a GATEWAY-35 destination vector allowing recombination of the HKT1-promoter sequence (Mäser et al., 2002)  produced a strong nuclear GFP signal in the cell layers surrounding the mature root xylem vessels, 7 primarily xylem parenchyma and pericycle ( Figure 1). GFP signal (green) in the root of Arabidopsis thaliana expressing pHKT1::NTF and pACT2::BirA. Cell walls were stained with propidium iodide (red) and images taken with the confocal microscope collecting fluorescent signals through 505-530 nm (GFP) and 560-615 (PI) nm filters after excitation at 488 nm. Zstacks were taken at 1 µm intervals and combined with Image J software to construct the images. A: Longitudinal section through the centre of the primary root. B: Orthogonal projection representing a transverse cross section of the primary root, reconstructed from the Z-stacks. XY: xylem, EP: epidermis, CX: cortex, EN: endodermis. Scale bars: 50 μm.

Methods). 12
Analysis based on GO-terms and keywords using DAVID (Huang et al., 2009) revealed 13 enrichment of annotations in the tagged cell types compared to all root transcripts ( Figure 3). 14 Annotation clusters related to transcriptional regulation had the highest scores with particularly 15 strong enrichment of transcription factors (TFs) that contain 'ethylene-responsive elements' 16 (AP/ERF). Another set of enriched annotations is linked to redox biology and iron, which could be 17 linked to ROS signaling in the tagged cell types. Additional significant clusters contain gene 18 functions related to auxin signaling, post-embryonic morphogenesis and oligopeptide transport as 19 would be expected given the role of the tagged tissues in lateral root development, xylem 20 differentiation and long-distance signaling. Enrichment of glucosinolate biosynthesis genes is 21 consistent with a role of the xylem in long-distance transport of glucosinolates (Andersen et al. 1 2013). Annotations related to photosynthesis were unexpected but could reflect perception of light 2 piped into the root via the xylem (Nimmo 2018). 3 4 Figure 3: Enrichment of functional gene annotations among genes with preferential expression in the xylem adjacent cell types. Enrichment scores were obtained using DAVID to compare representation of annotation terms in 2020 cell-type specific genes (≥ 4-fold transcript levels in tagged/non-tagged cell samples, p < 10 -4 ) with representation in all root expressed genes (25144 genes, see Methods). Names and statistics are only shown for one group in an enrichment cluster.

5
H3K27me3 distinguishes HKT1-expressing cell types from whole-root samples 6 As well as providing a new resource for mining root cell-type specific gene functions, the 7 dataset offered a starting point for investigating causal pathways underpinning and arising from 8 cell-type specific gene expression. To identify genes with differential histone modification patterns 9 in the pHKT1 tagged cell types we performed Isolation of Nuclei Tagged in Specific Cell Types  Figure 1). Each nuclear sample was further sub-divided and subjected to ChIP with 1 antibodies for H3, H3K4me3 or H3K27me3, followed by Illumina sequencing. The genome-wide 2 profiles of H3, H3K4me3 and H3K27me3 profiles in HKT1 and Whole Root were similar to those 3 previously reported for Arabidopsis roots (Sani et al., 2013). 4 For quantitative comparison we determined normalised cumulative ChIP-sequence reads over 5 the coding and immediate (200 bp) upstream sequence of each gene (see Methods). For 6 H3K27me3, PCA based on these values separated HKT1 and Whole Root samples along PC2 7 explaining 14% of the variance (Figure 4). One HKT1/Whole Root sample pair had produced lower 8 total read numbers than the other samples, and this technical difference dominated PC1. PCA 9 without these samples further unmasked the cell-type effect, now separating HKT1 and Whole Root 10 samples along PC1 explaining 72% of the variance (inset in Figure 4). We conclude that 11 H3K27me3 coverage of individual genes is a distinguishing feature of the root cell-types. By 12 contrast, PCA based on H3 or H3K4me3 levels did not separate HKT1 and Whole Root samples 13 (Supplemental Figure 2). A sixth sample only subjected to INTACT is also included. The nuclear isolates were subjected to ChIP with antibodies against H3K27me3. The PCA plot shown is based on cumulative H3K27me3 reads over gene bodies (see Methods). PC2 separates HKT1 samples from Whole Root samples explaining 14% of variation. Exclusion of one replicate pair with low total read number (insert) enhanced the distinction between cell types (PC1 explaining 72% of variation).

17
Genes with cell-type specific differences in H3K27me3 coverage have distinct functions 18 The distinction of the cell-types by PCA was based on small but consistent differences in gene 1 coverage with H3K27me3. To identify the differentially marked genes we used a sensitive ranking-2 based method (Rank Products;Breitling et al., 2004). 168 genes showed significantly lower 3 H3K27me3 and 263 genes showed significantly higher H3K27me3 coverage in HKT1 than in 4 Whole Root samples (FDR < 0.05). Very few genes showed significant differences in H3 levels (1 5 down and 19 up in HKT1 versus Whole Root,), and none of them showed also differential 6 H3K27me3. Therefore, the identified differences in H3K27me3 cannot be explained with a gain or 7 loss of nucleosomes. Similarly, very few genes had significant differences in H3K4me3 (6 up and 4 8 down in HKT1 versus Whole Root), and only two of them differed also in H3K27me3 (H3K4me3 9 up and H3K27me3 down). In summary, H3K27me3 was the most consistent difference between the 10 cell types but only affected a small percentage of the genome. 11 Based on annotation information for all differentially H3K27me3-marked genes enrichment 12 analysis with DAVID (Huang et al., 2009) revealed distinct functions of genes with cell-type 13 specific differences in H3K27me3 ( Figure 5). 14 15 Figure 5: Enrichment of functional gene annotations among genes with lower or higher H3k27me3 levels in xylem-adjacent cell types. Enrichment scores were obtained using DAVID. Gene sets tested contained genes with either significantly lower (Down) or significantly higher (Up) H3K27me3 levels. A similar sized reference set of genes with no changes showed no significant enrichment of functional annotations. Names and statistics are only shown for one group in an enrichment cluster.

16
Genes with lower H3K27me3 levels in HKT1 compared to Whole Root samples were enriched 17 for transcription factors, listed in Table 1, particularly those containing AP2/ERF domains. For 18 genes with higher H3K27me3 in HKT1 compared to Whole Root samples a significantly enriched 19 cluster contained annotation terms related to antimicrobial, antifungal, and defense functions. A 20 reference set of 200 genes randomly taken from genes with similar H3K27me3 levels in HKT1 and 1 Whole Root samples showed no significant enrichment (p < 10 -3 ) of functional annotations. 2 3 Differentially H3K27me3 marked genes have tissue-specific expression patterns 4 H327me3 is a repressive mark, and we therefore asked whether genes with lower H3K27me3 levels 5 had higher transcript levels in the tagged cell types. We consulted both previously published data 6 (Birnbaum et al., 2003;Brady et al., 2007) and our pHKT1-NTF based FACS dataset. Gene lists 7 based on previously employed root cell-markers yielded very few hits, emphasizing the fact that 8 these studies did not include specific markers for the root tissues where pHKT1 is active. Analysis  Figure 6. Overlap between genes with lower H3K27me3 levels and genes with preferential expression in the tagged cell types. Venn diagrams are based on lists of genes identified with FACS or INTACT. The genes with higher transcript level in tagged versus non-tagged cells (2020 genes, light blue circle) represent approximately 8% of the total transcriptome. The overlap between a random set of 200 genes with no change in H3K27me3 (pink circle) and the cell-type specific expressed genes (light blue circle) is 8% (16 genes). By contrast, 51% of genes with a cell-type specific decrease in H3K27me3 level (86 of 168 genes, dark blue circle) are preferentially expressed in the tagged cell types. Very little overlap (1 gene) was found with genes showing lower expression in tagged over non-tagged cell types (769 genes, brown circle).
Comparison of the 168 H3K27me3-depleted genes with our own FACS results identified a set 18 of 86 genes that had both significantly lower H3K27me3 and significantly higher transcript levels 19 in the tagged cell types than in the rest of the root ( Figure 6). The 51% overlap (86/168 genes) was 20 significantly higher (p=1.5x10 -44 ) than the 8% (2020/25144 genes) expected for any random subset 21 of genes. Indeed, overlap of the 2020 preferentially expressed genes with the 200 genes in the 22 reference set contained only 16 genes (8%) and overlap with the list of 263 genes with higher 23 H3K27me3 contained only 4 genes (1.5%). Only one of the 168 H3K27me3-depleted genes showed 1 a lower transcript level in tagged cells compared to non-tagged cells. 27 (64%) of the 42 2 H3K27me3-depleted transcription factors (Table 1) were preferentially expressed in the tagged cell 3 types, representing 31% of the 86 shared genes, which is significantly higher than the percentage of 4 TFs in the whole genome (8%), or among the 2020 cell-type specific genes (10%) with p-values of 5 5x10 -20 and 5x10 -18 respectively. 6 Overall, the analysis revealed a statistical association between cell-type specific loss of 7 H3K27me3 and increased gene expression, and a strong bias of the cell-type specific epigenetic de-8 repression for transcription factors. The 27 TFs (labelled 'HKT1' in Table 1) are therefore good 9 candidates for setting up local transcriptional networks. 14 of them were significantly upregulated 10 by a 24-h salt treatment of the roots (Figure 7). 11  Table 3 levels were determined by RT-qPCR. 14 of them showed significant induction by salt. pHKT1-NTF:pACT2-BirA plants were grown in hydroponics and treated with 150 mM NaCl for 24 hours (black bars), or not (open bars). Each biological replicate represents RNA harvested from 10 roots, grown in independent batches. Bars are means of transcript levels across 3 biological replicates, error bars are SE. Fold-changes and p-values (paired student's t test) are listed above bars. Within each biological replicate, transcript levels were normalised to reference gene YSL8 (At1g48370) in two technical replicates.

5
The cell-type specific transcription factor ZAT6 is de-methylated by REF6 6 H3K27me3 levels are determined by the relative rate of methylation and demethylation through 7 methyltransferases and demethylases, respectively. REF6 (At3g48430) encodes one of the 8 H3K27me3-specific demethylases in A. thaliana (Lu et al., 2011;Yan et al., 2018). Comparison of 9 the cell-type specifically de-repressed transcription factors (Table 1)   A: H3K27me3 levels (relative to Input) in regions 1 and 2 of ZAT6, and in the reference region, in roots of wildtype, ref6-3 and ref6-5 mutant lines as determined by anti-H3K27me3 ChIP-qPCR. Dots represent the results from three independently grown plant batches. Means are shown as orange lines. Different letters indicate significant differences (paired Student's t-test, p<0.05). B: ZAT6 transcript levels in roots of wildtype, ref6-3 and ref6-5 mutant lines as determined by RT-qPCR. Data are relative to the constitutively expressed YSL8. Dots represent the results from three independently grown plant batches. Means are shown as orange lines. Different letters indicate significant differences between genotypes (paired t-test, p<0.05).

Many cell-type specific transcripts are downstream of ZAT6 2
To test whether ZAT6 could establish a cell-type specific downstream network we analysed root 3 transcriptome of the two zat6 mutant lines by RNA-sequencing. PCA clearly distinguished between 4 mutant and wildtype samples based on normalized transcript levels ( Figure 10). Wildtype and 5 mutant samples were separated by PC1 explaining 78% of variance. The replicates of each mutant 6 line also grouped closely together, allowing distinction between the two lines primarily through 7 PC2 (explaining 5% of variance). 349 genes showed a significant decrease in transcript level (> 2-8 fold, p < 10 -3 ) in both zat6 mutant lines compared to wildtype. Comparison with the FACS dataset 9 revealed that most of the ZAT6-dependent genes were preferentially expressed in the tagged cell 10 types ( Figure 11).

3
Plotting for each gene the mean relative transcript level in zat6/WT against the relative 1 transcript level in tagged/non-tagged cell types shows a clear bias towards cell-type specific 2 expression among the ZAT6-dependent genes (Figure 11 A). 105 (30%) of the ZAT6-dependent 3 genes met the stringent cut-off criteria for preferential cell-type expression applied to the FACS 4 dataset (at least 4-fold, p < 10 -4 ), significantly more than the 8% expected by random (4x10 -33 ). We 5 conclude that cell-type specific de-repression of ZAT6 proliferates gene expression in a largely 6 tissue-delimited fashion thus establishing one branch of the local transcriptional network. 7 To verify that H3K27me3-demethylation of ZAT6 is critical for the expression of other genes in 8 this branch, we measured transcript levels of ZAT6-dependent cell-type specific genes (coloured in 9 Figure 11A) in ref6 mutants using RT-qPCR. All of the genes tested lost expression in the ref6 10 mutant lines ( Figure 11B). According to previous published data (Lu et al., 2011)  Legend for Table 1: 1 (2) Ratios of transcript levels between genotypes or cell types.

3
(4) Based on FACS RNA-seq analysis. T: Tagged sample (pHKT1-NTF), NT: Non-tagged sample 4 5 Functional annotation enrichment analysis of ZAT6-dependent genes with preferential 6 expression in the tagged root cell types identified a significant annotation cluster related to jasmonic 7 acid, wounding and defense signaling. The core gene sets underpinning this cluster contained 8 several JAZ proteins and oxylipin biosynthesis enzymes (Table 2). Transcription factors, 9 particularly those with AP2/ERF elements, were also again over-represented. 10 11 ZAT6 restricts sodium accumulatio in the shoots 12 Spatial co-expression of ZAT6 with HKT1, up-regulation by salt, and over-representation of ERFs 13 in the downstream network prompted us to investigate a possible role of ZAT6 in regulating long-14 distance transport of Na. Figure 12 shows Na and K concentrations in leaves of wildtype and zat6 15 plants after 3 days exposure to 100 mM NaCl. As expected, all salt-treated plants accumulated Na 16 in the shoots but Na concentrations were more variable and on average significantly higher in zat6 17 mutant plants than in wildtype plants. No differences were found for K. These results highlight a 18 function of ZAT6 in regulating root-to-shoot Na allocation under salt stress. 19 . Asterisks indicate significant difference between salt-treated zat6 and salt-treated wildtype (p < 0.01). All genotypes had significantly higher Na shoot concentrations in salt compared to control (p < 0.001).

HKT1 co-transcriptome and histone methylation patterns in xylem-adjacent cell types 2
The focus of this study was on cell types that control long-distance transport and communication 3 between the roots and the shoots, which is an essential prerequisite for adjusting shoot growth and 4 development to the presence of water, nutrients and stress factors in the soil environment. Previous 5 studies used root cell-type markers that are primarily expressed in the tips where the xylem cells are 6 still alive. However, the content of mature (dead) xylem vessels in the upper part of the root is 7 controlled by the adjacent cell layers. This was nicely shown by studies on the transporter HKT1. 8 HKT1 prevents accumulation of toxic Na + levels in the shoots by retrieving Na + from the foundation for further studies into specific gene functions in these tissues. All raw data have been 20 made publicly available for further data mining and specific gene lists with functional annotations 21 can be requested from the authors by interested parties. To facilitate characterization of cell-type 22 specific regulatory pathways we looked for potential primary regulators that do not require 23 upstream TFs for cell-type specific expression. In particular, we tested whether a cell-type specific 24 loss of H3K27me3 in a few genes could potentially underpin the entire cell-type specific 25 transcriptome. Integration of INTACT ChIP-Seq data with FACS RNA-Seq data identified a set of 26 genes with both lower H3K27me3 and higher expression in the HKT1-tagged cell types, highly 27 enriched for transcription factors (Table 1). Preferential targeting of transcription factors for cell-28 type specific changes in H3K27me3 was also found in guard-cell lineages (Lee et al., 2019) and 29 emerges as a candidate mechanism for establishing cell-type specific regulatory networks. 30

31
The ZAT6-branch of the HKT1 co-transcriptome 32 To test whether epigenetically de-repressed TFs could establish larger cell-type specific 33 transcriptional networks we further interrogated upstream and downstream regulation of the 34 transcription factor ZAT6. ZAT6 is preferentially expressed in the tagged cell-types and shows cell-35 type specific decrease of H3K27me3 (Table 1, Figures 8,9). Analysis of ref6 and zat6 mutants 1 showed that that (i) REF6 mediates H3K23me3 demethylation of ZAT6 ( Figure 8A, 9A), (ii) REF6 2 is required for expression of ZAT6 ( Figure 9B), (iii) a large number of genes are transcriptionally 3 dependent on ZAT6 (Figure 10), most of which show cell-type specific expression (Fig. 11A), and 4 (iv) REF6-dependence of ZAT6 expression is levied onto downstream gene that are not direct 5 REF6-targets ( Figure 11B). These results provide proof-of-concept that cell-type specific epigenetic 6 de-repression of a single transcription factor can establish cell-type specific expression of a large 7 number of genes. Supplemental Figure 3 summarizes this conclusion in a working model. 8 9

What is upstream of ZAT6 de-repression? 10
But why is H3K27me3 removal from ZAT6 cell-type specific? Our FACS experiments showed that 11 REF6 itself is not preferentially expressed in the tagged cell-types. ZAT6 is a candidate target of 12 H2A ubiquitination, which can recruit REF6 (Kralemann et al., 2020), Table S5

Downstream 'proliferation' of cell-type specific de-repression of ZAT6 21
Assessing cell-type specificity of ZAT6 downstream genes through RNA-Seq of zat6 mutants 22 allowed us to quantitatively interrogate the conservation of spatial expression within transcriptional 23 pathways. We found that at least a third of the ZAT6-dependent transcriptome maintain strong cell-24 type preference. Preferential expression in the tagged cell types is increasingly lost among genes 25 that are weakly dependent on ZAT6 ( Figure 11A

ZAT6-dependent gene functions in the xylem adjacent cell layers 32
Previous studies have reported roles of ZAT6 in chilling, osmotic stress, salt, cadmium, 33 phosphate and pathogen responses as well as flowering (Devaiah et al., 2007;Shi et al., 2014;Chen 34 et al., 2016;Liu et al., 2013). However, this evidence was derived from analysing whole seedlings 35 and shoots of zat6 mutants, or from ectopic 35S-driven over-expression of ZAT6, and is therefore 1 not informative for cell-type specific function in roots. Expression of the glutathione biosynthesis 2 gene GSH1 (At4g23100), which was identified as a direct target of ZAT6 in heterologous systems 3 and linked to Cd-tolerance, was not ZAT6-dependent in roots. However, two ZAT6-dependent 4 heavy metal-chelating/transport proteins are specifically expressed in the xylem-adjacent cells 5 (AT5G05365, AT5G52760) and should be tested for a role in long-distance transport. 6 ZAT6 expression has been reported to be up-regulated by biotic and abiotic stress (Devaiah et 7 al., 2007;Shi et al., 2014;Chen et al., 2016;Liu et al., 2013) and we measured up-regulation in the 8 roots by salt (Figure 7). Furthermore, salt-treated zat6 mutant plants hyperaccumulated Na + in the 9 shoots compared to wildtype (Figure 12). This phenotype mimics that of mutants for ion 10 transporters, such as HKT1 and SOS1 (Shi et al., 2002, Møller et al., 2009) that remove Na + from 11 the root xylem. The ZAT6-dependent transcriptome did not include any known Na + transporter 12 genes, but it contained candidates for their posttranslational regulation. Notably, the type-2C protein 13 phosphatase PP2C49 (At3g62260), previously shown to regulate HKT1 (Chu et al. 2020), was 14 preferentially expressed in the tagged cell types and lost expression in the zat6 mutants. Our study 15 provides a limited set of spatially co-expressed genes for systematic interrogation of regulatory 16 pathways through reverse genetics. 17 The functional enrichment analyses of cell-type specific genes that showed both H3K27me3 de-18 methylation and ZAT6-dependency also pinpointed hormonal signals . While not all AP/ERF TFs 19 are bona fide targets of ethylene, their over-representation ties in with reports on radial ethylene 20 signals controlling xylem content and long-distance transport of Na + and K + (Jiang et al., 2013). A 21 second significantly enriched annotation cluster pointed to ZAT6 mediating cell-type specific 22 jasmonate responses (Table 2) and supports a role of xylem-adjacent cells in communicating 23 wounding and disease in the root to the shoot (Biere and Goverse, 2016). 24

25
In summary, our study provides proof-of-concept for establishment of a cell-type specific 26 transcriptional network from an epigenetically regulated TF. Alongside ZAT6, 26 TFs with both 27 cell-type specific expression and cell-type specific H3K27me3-demethylation were identified 28 (Table 1). Assuming a similar-sized 'followership' as for ZAT6, they could generate cell-type 29 specific transcriptome of over 2000 genes thereby offering a handle to systematically delineate 30 regulatory modules in root xylem-surrounding cell types from the 'bottom up'. Understanding the 31 regulatory networks in a strategic location at the root/shoot interface is essential for improving 32 water and nutrient allocation in plants as well as resilience against soil-borne diseases and toxins.

Confocal microscopy for GFP localization 23
A LSM 510 confocal microscope (Carl Zeiss, Jena, Germany) was used to visualise GFP 24 fluorescence in roots after staining of cell walls with propidium iodide (10 µg/mL propidium iodide 25 for 10 min). The excitation wavelength for both GFP and propidium iodide was 488 nm. 26 Fluorescent signals were collected through 505-530 nm and 560-615 nm filters, respectively. Z-27 stacks were taken at 1 µm intervals and used to re-construct transverse cross sections with Image J. 28 29

RNA-seq data analysis 28
The raw fasta files were pre-processed to trim the 3' end adapter and remove low quality 3' bases 29 (with a PHRED score less than 15) with cutadapt (version 1.5) (Martin, 2011). Reads trimmed to 30 less than 54 bases in length were also removed. Transcript expression quantification was performed expressed genes. Chloroplast and mitochondrial genes, and all genes with an overall base mean < 35 1.5, were excluded from further analysis, resulting in a total number of 25144 'root expressed' 1 nuclear genes. For identification of genes with preferential expression in the pHKT-NTF tagged (T) 2 versus non-tagged (NT) cell types we applied a cut-off of T/NT > 4-fold and p < 10 -4 . Annotation 3 enrichment analysis was performed using DAVID (Huang et al., 2009). 4

Chromatin immunoprecipitation 22
Immediately after nuclei purification, formaldehyde was added to both 'Whole Root' and 'HKT1' 23 samples to a final concentration of 1% (v/v) for cross-linking on ice. After quenching nuclei from 24 the 'Whole Root' sample were pelleted, while bead-bound nuclei from the 'HKT1' sample were 25 captured with a magnetic rack. Chromatin was sonicated to a size of approx. 400 bp and fraction 26 kept as 'Input'. Antibodies against H3 (Abcam, #ab1791), H3K4me3 (Diagenode, #C15410003) or 27 H3K27me3 (Diagenode, #C15410069) were added for overnight immunoprecipitation. The 28 chromatin-antibody complex was recovered with protein A Dynabeads (Invitrogen) and washed. 29 DNA was eluted, reverse cross-linked and treated with proteinase K. DNA was purified using 30 MinElute (Qiagen). Successful ChIP was confirmed through a quality control as reported before 31 (Sani et al., 2013). Enrichment of sequences known to be associated with H3K27me3 (positive 32 control, At5g56920) or not (negative control, At5g56900) was determined by qPCR using the Input 33 sample for normalization. Primers used are listed in Supplemental Table 1.  Table 2. For RT-qPCR, YSL8 30 (At1g48370) was used as reference. Primers for ChIP qPCR were designed to amplify two regions 31 in ZAT6 (At5g04340) and a reference region (At5g04410). Values were normalised to Input. 32 33

Measurement of shoot ion content 34
For analysis of shoot ion contents, plants were transferred from plates to hydroponics at 20 days 1 after germination and grown on control media for another week. 100 mM NaCl was supplied (or 2 not, control) in independent treatments of each plant. Whole rosettes were harvested 120 hours after 3 treatment. Tissues were dried at 60 ℃ and dry weights determined before incubation in 5 ml 1 M 4 HCl for 48 hours. Supernatants of extract were diluted (20-100x), and Na and K concentrations 5 determined in a flame photometer (410, Sherwood-Scientific Ltd, UK) based on NaCl or KCl 6 standard curves. Concentrations were calculated back to undiluted extract and normalized to shoot 7 dry weight of each plant.  and fluorescent signals for DAPI (blue staining of nuclei) and GFP (green signal of nuclear 7 envelope). Scale bar is 10 µm. B: Relative ratio of Ler to Col-0 DNA as determined by qPCR using 8 the 'spike-in' method (primers listed in Supplemental Table 1). Snp: spike-in non-purified sample; 9 Sp: spike-in purified sample. Bars represent means and standard errors of all the purifications 10 carried out for this study (n=8). Biotin-labelled nuclei (HKT1, black symbols) were isolated from roots of pHKT1::NTF 6 pACT2::BirA lines using pulldown with streptavidin (INTACT). Total nuclear preparations (not 7 subjected to INTACT) from the same root material represent all cell types (Whole Root, open 8 symbols). The nuclear isolates were subjected to ChIP with antibodies against H3K27me3. 9 Replicate samples were obtained from independently grown plant batches. In this model a relatively small number of core regulators such as ZAT6 are cell-type specifically 6 de-repressed during differentiation through the activity of H3K27me3 demethylases such as REF6 7 (K27 indicates H3K27me3). Regulation of multiple genes by each core regulators (e.g. solid lines 8 for ZAT6-dependent genes) results in downstream proliferation of the expression pattern and 9 establishes cell-type specific regulatory networks. The model proposes a primary mechanism for 10 cell-type specific de-repression, which could be further enhanced through feedback regulation at the 11 epigenetic level (brown arrow) and the transcriptional level (blue arrow), involving for example 12 cell-type specific expression of chromatin modifiers, generation of cell-type specific signals and