Transcription-driven Chromatin Repression of Intragenic Promoters

Progression of RNA polymerase II (RNAPII) transcription relies on the appropriately positioned activities of elongation factors. The resulting profile of factors and chromatin signatures along transcription units provides a “positional information system” for transcribing RNAPII. Here, we investigate a chromatin-based mechanism that suppresses intragenic initiation of RNAPII transcription. We demonstrate that RNAPII transcription across gene promoters represses their function in plants. This repression is characterized by reduced promoter-specific molecular signatures and increased molecular signatures associated with RNAPII elongation. The FACT histone chaperone complex is required for this repression mechanism. Genome-wide mapping of Transcription Start Sites (TSSs) reveals thousands of discrete intragenic TSS positions in FACT mutants. Histone 3 lysine 4 mono-methylation poises exonic sites to initiate RNAPII transcription in FACT mutants. Uncovering the mechanism for intragenic TSS repression through the act of RNAPII elongation has important implications for understanding pervasive RNAPII transcription and the regulation of transcript isoform diversity.


Summary:
Progression of RNA polymerase II (RNAPII) transcription relies on the appropriately positioned activities of elongation factors. The resulting profile of factors and chromatin signatures along transcription units provides a "positional information system" for transcribing RNAPII. Here, we investigate a chromatin-based mechanism that suppresses intragenic initiation of RNAPII transcription. We demonstrate that RNAPII transcription across gene promoters represses their function in plants. This repression is characterized by reduced promoter-specific molecular signatures and increased molecular signatures associated with RNAPII elongation. The FACT histone chaperone complex is required for this repression mechanism. Genome-wide mapping of Transcription Start Sites (TSSs) reveals thousands of discrete intragenic TSS positions in FACT mutants. Histone 3 lysine 4 mono-methylation poises exonic sites to initiate RNAPII transcription in FACT mutants. Uncovering the mechanism for intragenic TSS repression through the act of RNAPII elongation has important implications for understanding pervasive RNAPII transcription and the regulation of transcript isoform diversity.

Introduction:
Plasticity at the beginning and end of transcripts multiplies the RNA species that can be generated from genomes. RNA results from RNAPII activity at genes, but also from pervasive transcription of abundant non-coding genomic regions (Jensen et al., 2013;Mellor et al., 2016). Pervasive transcription may result in overlapping transcripts, for example by initiating intragenic transcription leading to the production of alternative transcript isoforms (Davuluri et al., 2008). Alternative Transcription Start Sites (TSSs) expand RNA isoform diversity, may result in functionally different RNA and proteins specific to disease, and allow for multiple transcriptional inputs from a single gene (Arner et al., 2015;Wiesner et al., 2015). However, the mechanisms of alternative TSS activation, repression and regulation are poorly understood in higher eukaryotes.
Repression of a gene promoter by overlapping RNAPII transcription was originally described for two tandemly arranged human α-globin gene copies (Proudfoot, 1986). Read-through transcription from the upstream α-globin gene positions the downstream promoter in the middle of transcription unit spanning both gene copies. The repression of the downstream promoter through the act of RNAPII transcription is referred to as Transcriptional Interference (TI) (Ard et al., 2017). The core of this mechanism relies on the progression of RNAPII transcription through distinct stages (Buratowski, 2009). Each stage is characterized by the co-transcriptional recruitment of factors involved in nascent RNA processing and chromatin modifications (Li et al., 2007). Dynamic phosphorylation of residues in the C-terminal YSPTSPS repeat region of the largest RNAPII subunit coordinates the progression through transcription cycle by recruiting stage-specific factors (Corden, 2013;Eick and Geyer, 2013).
Metagene analyses of stage-specific transcription factors and chromatin signatures in diverse organisms strikingly visualizes many common changes associated with RNAPII progression from the beginning to the end of active transcription units (Descostes et al., 2014;Gerstein et al., 2010;Hajheidari et al., 2013;Kharchenko et al., 2011;Mayer et al., 2010;Pokholok et al., 2005). These signatures provide a "positional information system" (POINS) for RNAPII to coordinate molecular events required for each stage of transcription (Buratowski, 2009).
Co-transcriptional chromatin signatures are common across species, yet their roles in the regulation of intragenic TSSs often awaits experimental validation.
Many factors characterizing POINS are active in plants (Hajheidari et al., 2013;Van Lijsebettens and Grasser, 2014). The Arabidopsis FACT complex is physically associated with multiple RNAPII elongation factors, chromatin modifiers and elongation specific RNAPII isoforms (Antosz et al., 2017;Duroux et al., 2004). Reduced FACT activity results in developmental defects (Lolas et al., 2010) linked to imprinting (Ikeda et al., 2011), yet the role of FACT in TSS selection in plants is unclear. Genome-wide TSS mapping in Arabidopsis suggest that a choice between alternative TSSs exists for most transcripts . Protein isoform diversity control in response to light through regulated of TSS choice underpins the biological significance of this mechanism (Ushijima et al., 2017). TSS choice may also regulate gene expression at the level of translation by the inclusion of an upstream open reading frame (uORF) (von Arnim et al., 2014).
Despite the functional significance of alternative TSS choice little is known about the molecular mechanisms regulating this phenomenon in plants.
Here, we demonstrate the repressive effect of RNAPII elongation across gene promoters in Arabidopsis. We identify chromatin and RNAPII signatures associated with this form of gene regulation by "repressive transcription" that relies on chromatin remodeling by the FACT complex. We identify several thousand FACT-sensitive intragenic TSSs, revealing a role for FACT in preventing initiation of RNAPII transcription from within transcription units. Our study characterizes the molecular events involved in repressing RNAPII initiation by the process of RNAPII elongation for the first time in the context of a multicellular organism.

Gene promoter repression by upstream RNAPII transcription in Arabidopsis thaliana
To investigate gene repression through the act of RNAPII transcription across promoter regions in higher organisms, we performed a literature screen of Arabidopsis T-DNA insertion mutants with lossof-function phenotypes (Lloyd and Meinke, 2012). This specific type of T-DNA mutants must: 1.) be inserted upstream of gene TSSs, 2.) show read-through transcription into downstream genes, and 3.) segregate as recessive loss-of-function phenotype. Application of these criteria identified the quasimodo1-1 (qua1-1) and red fluorescence in darkness 1-1 (rfd1-1) mutants as candidate mutants for further analysis (Bouton et al., 2002;Hedtke and Grimm, 2009). QUASIMODO1 (QUA1) encodes a glycosyltransferase required for the biosynthesis of celladhesion promoting pectins (Bouton et al., 2002). The qua1-1 T-DNA mutation is inserted 117 bp upstream of the annotated translational start site ( Fig. 1A; SFig. 1A). The cell-adhesion defect in qua1-1 results in dwarfed growth and ruthenium red staining of dark grown qua1-1 hypocotyls (Fig. 1B). We detect elevated QUA1 expression in qua1-1 compared to wild type by RT-qPCR (Fig. 1C). Northern blotting reveals an abundant T-DNA-QUA1 compound transcript in qua1-1 instead of the QUA1 mRNA ( Fig. 1D) (Bouton et al., 2002). The extended transcript detected in qua1-1 corresponds to a predicted transcript initiating within the T-DNA and extending into the downstream QUA1 gene (SFig. 1B). RED FLUORESCENT IN DARKNESS 1 (RFD1) encodes RibA1, the first enzyme in the plant riboflavin biosynthesis pathway (Hedtke and Grimm, 2009). The T-DNA insertion is located 307 bp upstream of the RFD1 translation start ( Fig. 1E; SFig. 1C). Under standard light conditions, soil-grown homozygous rfd1-1 mutants die as white cotyledons (Fig. 1F) (Hedtke and Grimm, 2009). However, we are able to grow homozygous rfd1-1 mutants to seed under low light conditions, enabling comparative analysis of the RFD1 transcript pattern in wild type and homozygous rfd1-1 mutants.
Our transcript analyses in qua1-1 and rfd1-1 are consistent with the hypothesis that initiation from the downstream gene promoter is repressed through the act of RNAPII transcription.
To test if the genomic region between the rfd1-1 T-DNA insertion and the translation start of RFD1 can function as a promoter (designated as TIpRFD1, Fig. 1E; SFig. 1C) we assayed marker gene expression. We first monitored GUS-staining following transient expression in Nicotiana benthamiana and Arabidopsis leaves ( Fig. 2A,B). We detect GUS activity in TIpRFD1-GUS compared to the p19 control injection in N. benthamiana, but it is weaker than the p35S-GUS positive control. We also detect stronger GUS staining and eYFP activity compared to p19 negative control injections in Arabidopsis (Fig. 2B, SFig. 2). To test if TIp can drive gene expression in relevant tissues and at sufficiently high levels, we performed a molecular complementation of the read-through mutants with genomic constructs driven by their respective TIp. We detect RFD1-FLAG protein expression to varying degrees in independent transformant lines by western blotting (Fig. 2C). Importantly, RFD1 expression driven by TIpRFD1-RFD1-FLAG complements the rfd1-1 phenotype (Fig. 2D). Likewise, we detect QUA1-FLAG protein expression in independent TIpQUA1-QUA1-FLAG transformant lines by western blotting, and these lines complement the qua1-1 phenotype (Fig. 2E,F). Thus, TIp DNA regions provide necessary and sufficient promoter activity to drive functional RFD1 or QUA1 expression. Interfering RNAPII transcription across TIp provides a candidate mechanism to explain the repression of initiation despite transcriptional activity at these regions.

Elevated RNAPII elongation signatures are found at promoters repressed through the act of upstream RNAPII transcription in Arabidopsis
Repressive RNAPII elongation across TIp in qua1-1 and rfd1-1 mutants may impact on molecular signatures associated with RNAPII elongation and initiation at TIp. To test this, we performed quantitative chromatin immunoprecipitation (qChIP) experiments for RNAPII initiation and elongation hallmarks. The elongating form of RNAPII (RNAPII-Ser2P) is enriched towards the 3' end of the QUA1 gene and depleted from the QUA1 promoter in wild type Arabidopsis (SFig. 3A, B). H3K36me3 is enriched towards the 5' end of genes in Arabidopsis, while H3K36me2 corresponds to the elongation phase and accumulates towards the 3' end (Mahrez et al., 2016). We find the same pattern along the QUA1 gene (SFig. 3C-D). Histone modifications of active promoters such as histone H3 acetylation (H3ac) and H3K4me3 are enriched towards the QUA1 promoter (SFig. 3E-F) (Mahrez et al., 2016;Roudier et al., 2011;Zhang et al., 2016a). RNAPII initiation and elongation can be distinguished by our qChIP analysis.
We profiled qua1-1 and rfd1-1 mutants by qChIP to determine the impact of upstream RNAPII transcription across TIpQUA1 and TIpRFD1. Compared to their respective wild type ecotype, significantly higher levels of RNAPII-Ser2P were present at the position of promoter-proximal primer pairs in qua1-1 and rfd1-1 ( Fig. 3A and 3B). These results support increased RNAPII elongation across the downstream promoter. Since bulk histone density remains largely unchanged across QUA1 and RFD1 in their respective mutants (SFig. 3G-I), we tested the presence of the elongation-specific chromatin signature H3K36me2. The mutants displayed increased levels of H3K36me2 at TIpQUA1 and TIpRFD1 ( Fig. 3C and 3D). The increase of RNAPII elongation signatures at these promoters during repression indicates that these regions may now identify as zones of RNAPII elongation instead of promoters.
Collectively, these results demonstrate that upstream RNAPII transcription shifts the POINS to specify downstream promoters as intragenic regions. Our data suggest that promoter repression in these mutants is driven by transcription-mediated changes to promoter chromatin status.

Arabidopsis FACT is required for gene repression through the act of upstream RNAPII transcription
Our analyses support that gene promoters can be repressed by interfering RNAPII elongation in Arabidopsis. We predict that factors associated with RNAPII elongation, such as the FACT complex, may be required for repression. To test the role of FACT in promoter repression by read-through transcription in Arabidopsis, we combined the previously described knock-down alleles of spt16-1 and ssrp1-2 mutants with qua1-1 (Lolas et al., 2010). Ruthenium red staining comparing single and double mutants revealed patches of unstained hypocotyls in spt16-1 qua1-1 compared to qua1-1 (Fig. 4A).
Presumably this can be explained by a stronger knock-down of protein levels in ssrp1-2 compared to spt16-1 (Lolas et al., 2010). To test the link between H3K36me and FACT we assayed genetic interactions between qua1-1 and a mutation in the Arabidopsis H3K36me methyltransferase SDG8/ASHH2 (Grini et al., 2009;Zhao et al., 2005). Interestingly, we find no evidence for suppression of qua1-1 by sdg8-2 (SFig. 4). Together, our data demonstrate that the FACT complex is required for the cell adhesion defects observed in qua1-1.
If phenotypic suppression of qua1-1 through fact mutants was mechanistically linked to gene repression through the act of upstream interfering transcription, we would predict transcriptional changes. To examine the pattern of QUA1 transcripts we performed northern blotting in single and double mutants. While the transcript pattern in spt16-1 and ssrp1-2 is indistinguishable from wild type controls, we observe new transcript patterns in spt16-1 qua1-1 and ssrp1-2 qua1-1 double mutants compared to qua1-1 (Fig. 4C). Importantly, we detect a reduced size of the interfering transcript in fact mutants, arguing for a 5'-truncated transcript initiating from a cryptic TSSs. Notably, the wild type QUA1 mRNA is not restored. This indicates that functional mRNAs are produced from an upstream promoter accessible in fact mutants. Our results support the conclusion that the activity of the FACT complex as part of RNAPII elongation suppresses intragenic TSSs.

Arabidopsis FACT restricts the activity of intragenic promoters
To test if FACT suppresses endogenous intragenic TSSs, we measured Arabidopsis TSSs by 5'-CAPsequencing (TSS-seq) (Pelechano et al., 2016). We called 77292 TSS clusters and annotated them by genomic location. Many TSS clusters (n=31013, or 40.1%) mapped to gene promoters (SFig. 5A). We obtained on average 47 million raw reads for two biological repeats of wild type, spt16-1, and ssrp1-2 (Supplementary Table S1). Biological repeats showed a high degree of correlation (SFig. 5B). TSSseq also revealed reduced TSS signals at SPT16 and SSRP1 genes in the respective mutants consistent with reduced expression (SFig. 5C-D). We examined the overlap of our TSS clusters with TSSs identified by CAGE (Cap Analysis Gene Expression) . 74.8% of TSS clusters in core promoters overlap with at least one previously reported CAGE peak (SFig. 5E and Supplementary Table S2), indicating very good overlap across techniques and samples. Alternative mRNA isoforms of the At4G08390 gene are differentially targeted to mitochondria or chloroplast, and our data resolves TSSs corresponding to these isoforms (SFig. 5F) (Obara et al., 2002). Interestingly, our TSS-seq data reveals almost as many TSSs in exons (n=30831, or 39.9%) as in gene promoters  Table S3). In conclusion, these data illustrate high reproducibility of our TSS-seq methodology, and its abilities to validate TSSs as well as to reveal novel TSSs.
To test the role of FACT in regulating TSSs in Arabidopsis, we divided the TSSs into three groups: i) constitutive TSSs detected in both wild type and FACT mutants (n=60966, or 78.9%); ii) wild-type specific TSSs (n=669, or 0.87%); and iii) TSSs specifically detected in fact mutants (n=15657, or 20.3%, Supplementary Table S4). The 23.4-fold increase of FACT-specific TSsS over the wild-type specific TSSs suggests that the FACT complex largely represses TSSs. FACT-specific TSSs map to intragenic locations, particularly exons (Fig. 5A). However, TSSs induced in fact mutants have a lower TSS-seq count compared to the basal TSS set indicating lower expression ( Fig.   5B-C). The large majority of FACT-specific TSSs (7460 out of 9093, or 82%) were detected in both mutants (Fig. 5D). In addition, FACT-specific TSSs overlap less frequently with TSSs identified by CAGE (SFig. 5E and Supplementary Table S2). As much as 80.7% of FACT-specific exonic TSS clusters did not overlap with any CAGE peak. The At5G18500 gene illustrates the induction of an intronic TSS in fact mutants (Fig. 5E). The AT4G15530 pyruvate orthophosphate dikinase (PPDK) gene reveals preferential usage of a downstream TSS in fact mutants (Fig. 5F). While the longer transcript localizes to chloroplasts, the shorter PPDK isoform targeted to the cytoplasm instead (Parsley and Hibberd, 2006). The AT1G10120 gene TSS selection demonstrates selective RNAPII initiation from a downstream 5'UTR site in fact mutants (Fig. 5G). The number of FACT-sensitive intragenic TSSs argues for high selectivity and tight regulation of intragenic regions functioning as TSSs. Our TSS-seq data therefore support a role for the FACT complex as part of POINS in Arabidopsis, with a specific function in suppressing intragenic cryptic TSSs.

H3K4me1 in wild type marks intragenic regions that function as TSSs in FACT mutants
Common signatures in DNA sequence or chromatin environment may predispose intragenic regions to function as TSSs in fact mutants. We tested differential DNA-motif enrichment in exonic FACT-specific TSS clusters compared to basal exonic TSS. However, we detect no differentially enriched sequence motif or position bias (SFig. 5G). We next compared the chromatin signatures around exonic FACTspecific TSSs using available data for Arabidopsis in bud and leaf tissue samples (Zhang et al., 2016b). We compared chromatin signatures centered on five sets of genomic locations: FACT-specific exonic TSS, exonic TSSs not regulated by FACT (basal TSSs), promoter TSSs, random exonic positions and random genomic positions. Strikingly, we find the highest H3K4me1 enrichment at exonic sites in wild-type for FACT-specific TSSs ( Fig. 6A-B). These regions also show less H3K27me3 compared to random exonic TSSs or random genomic locations ( Fig. 6C-D). However, H3K27me3 is also reduced at basal exonic TSSs. If exonic TSS positions repressed by FACT were characterized by reduced chromatin condensation we would predict these sites to be particularly sensitive to DNaseI. Our analyses of DNA elements and chromatin signatures link H3K4me1 to exonic regions poised to function as TSSs in fact mutants. Our results are consistent with a chromatin-based mechanism predisposing exonic sites as cryptic TSSs controlled by the FACT complex.
All in all, our data support a model in which FACT is required for the repression of intragenic cryptic TSSs (Fig. 7). Read-through transcription can re-define gene promoters as intragenic to bring them under the influence of POINS. Repression of promoter function relies on a loss of initiationspecific RNAPII hallmarks, and a gain of elongation-specific signatures ( Fig. 7A-B). The FACT complex represses intragenic transcription and maintains associated chromatin signatures (Fig. 7C).
We identify a large number of exonic TSSs repressed throughout the Arabidopsis genome that are predisposed to function as TSSs by H3K4me1 in the absence of FACT-mediated repression. Our findings describe a conserved co-transcriptional chromatin-based mechanism shaping gene regulation and transcript isoform diversity.

Discussion:
TSSs shape RNA isoform expression but little is known about the mechanisms regulating TSS choice within transcription units. RNAPII transcription across gene promoters may re-define gene promoters as "intragenic" and repress them by mechanisms inhibiting initiation from within transcription units. We leveraged Arabidopsis T-DNA read-through mutants to identify a role of the conserved FACT histone chaperone complex in the repression of intragenic TSSs in a multicellular organism. Consistently, we identify a large number of intragenic TSSs controlled by FACT, particularly from exons with the chromatin signature H3K4me1.
Three activities of the FACT complex may explain a key role in repressing intragenic TSSs across species are: 1.) stimulation of RNAPII elongation, 2.) histone re-assembly in the wake of RNAPII transcription, and 3.) recycling of old histones to maintain POINS.
First, FACT stimulates RNAPII transcription of DNA templates packaged in nucleosome structures (Belotserkovskaya et al., 2003). Structural analyses suggest that the FACT complex directly binds nucleosomes on several contacts of histone proteins, stabilizing otherwise energetically unfavorable nucleosome conformations that weaken nucleosome binding to DNA (Hondele and Ladurner, 2013).
Stabilization of partly unfolded nucleosome intermediates facilitates RNAPII progression through nucleosome barriers. Defective FACT may result in "transcription stress" through stalled or arrested RNAPII molecules in transcription units. Transcription stress serves as a DNA damage signal associated with proteolytic degradation of stalled RNAPII (Wilson et al., 2013). DNA damageassociated chromatin changes facilitate the initiation of RNAPII transcription that could help to explain elevated TSSs in FACT mutants (Francia et al., 2012;Price and D'Andrea, 2013). RNAPII elongation defects may hence trigger RNAPII initiation.
Second, FACT aids the re-assembly of nucleosomes in the wake of transcribing RNAPII (Belotserkovskaya et al., 2003;Schwabish and Struhl, 2004). Consistently, reduced nucleosome coverage within transcription units has previously been reported in human and yeast FACT mutants (Carvalho et al., 2013;van Bakel et al., 2013). Nucleosome Depleted Regions (NDRs) are associated with TSSs at gene promoters and may trigger firing from cryptic intragenic TSSs observed by us and others (Carvalho et al., 2013;Feng et al., 2016;Kaplan et al., 2003). However, we find no predisposition in wild type for relaxed chromatin packaging at exonic regions that function as TSSs in FACT mutants (Fig. 6). While it remains likely that NDRs are associated with intragenic TSSs in FACT mutants, these data suggest that NDRs would result from reduced FACT histone re-assembly activity rather than sequences of chromatin signatures destabilizing nucleosomes.
Third, the propensity of FACT to deposit "old histones" in the wake of RNAPII transcription represents an intuitive mechanism to maintain the co-transcriptional positional information provided by chromatin signatures (Jamai et al., 2009). Consistently, defective FACT disrupts POINS as is evidenced by the incorporation of the promoter-enriched histone variant H2A.Z within transcription units in FACT mutants (Jeronimo et al., 2015). Future studies will be required to dissect the individual and cumulative contributions of defects in RNAPII elongation, nucleosome re-positioning, and POINS establishment in the up-regulation of intragenic TSSs observed in FACT mutants.
Promoter-proximal histone acetylation is associated with active promoters. Histone deacetylases associate with elongating RNAPII to repress the activity of intragenic TSSs (Carrozza et al., 2005;Keogh et al., 2005). Our findings of reduced histone acetylation in promoter regions in qua1-1 and rfd1-1 read-through mutants support this observation in plants (Fig. 3). We find no evidence for a predisposition to increased H3K27ac at exonic regions in wild type that function as FACT-specific TSSs (Fig. 6). These data argue that reduced histone acetylation within gene-bodies results from the process of RNAPII elongation, although different specificities of histone acetylation antibodies could also contribute to this observation. Proteomic screens identified several histone deacetylases (HDACs) associating RNAPII elongation complexes in Arabidopsis (Antosz et al., 2017). It remains to be tested which complexes provide HDAC activity linked to the suppression of intragenic TSSs.
H3K36me3 correlates with RNAPII elongation and linked suppression of intragenic TSSs in mammals and yeast. However, H3K36me3 is a promoter-proximal chromatin signature in Arabidopsis, while H3K36me2 co-localizes with RNAPII elongation (Mahrez et al., 2016). Consistently, our chromatin state analyses in qua1-1 and rfd1 support H3K36me2 as a chromatin signature of RNAPII elongation, and H3K36me3 as a signature associated with initiation zones near promoters. Interestingly, we find no evidence for a role of the Arabidopsis H3K36 methyltransferase SDG8/ASHH2 is gene repression through the act of RNAPII transcription in qua1-1 (SFig. 4). Perhaps one of the other 47 SET-Domain Genes (SDGs) in Arabidopsis contributes a redundant activity to the repress of intragenic TSSs (Ng et al., 2007). Future research will address the question if H3K36me2 fulfils the role of H3K36me3 in plants, or if mechanisms repressing intragenic TSSs independent of H3K36me exist. Our screen for chromatin signatures characterizing exonic regions poised to function as TSSs in FACT mutants identifies histone 3 lysine 4 mono-methylation (H3K4me1). The association of H3K4me1 with enhancer activity has recently been called into question (Dorighi et al., 2017), however H3K4me1 is indicative of RNAPII transcription. In Arabidopsis, intragenic H3K4me1 counteracts H3K9me2mediated chromatin repression (Inagaki et al., 2017). High levels of H3K4me1 at FACT-respressed TSS positions may facilitate the transition towards H3K4me3 and promoter function of exonic regions.
Further dissection of the regulation of intragenic promoters in plants promises to reveal novel mechanisms for the co-transcriptional HDAC recruitment, involved histone methyltransferases and novel components of chromatin-based suppression of intragenic TSSs.
Alternative TSS selection provides a mechanism to generate transcript isoform diversity from a given genome. Light signaling triggers the choice between alternative TSSs in Arabidopsis, with profound effects on subcellular protein targeting (Ushijima et al., 2017). While a role of FACT in plant light signaling remains to be firmly established, it is conceivable that FACT contributes to the regulation of light-sensitive intragenic TSS selection. TSSs at gene promoters identified by our TSSseq approach agrees well with promoter TSSs identified by CAGE, despite the fact that CAGE data used for comparison was sampled using different developmental stages .
However, our TSS-seq data in FACT mutants identify thousands of novel TSSs mapping to intragenic regions, highlighting that FACT activity likely protects plant genomes from intragenic transcription initiation. While our bioinformatic methods to detect intragenic TSSs are rigorous, and we identify a large overlap between TSSs in mutants of two FACT subunits, it is clear that expression of FACTspecific TSSs is low compared to basal TSSs (Fig. 5). Since null mutations in FACT are not viable, we characterized knock-down alleles. This suggests we are underestimating the global impact of FACT activity on intragenic TSS suppression, both in in quality and quantity (SFig. 5) (Lolas et al., 2010).
Moreover, an low expression of fact-specific TSSs may be explained by targeted RNA degradation, as has been shown for cryptic expression of divergent lncRNA transcripts (Ntini et al., 2013). While specific examples remain to be characterized in plants, we demonstrate that the underlying mechanism of repressive RNAPII transcription is operational.
Future research will be required to fully address potential regulatory functions of intragenic TSSs. Our study offers a platform to query the role of intragenic TSSs in plant signalling and development, both by providing conceptual advances for a mechanism regulating intragenic TSSs, and by the identification of thousands of exonic sites poised to function as TSSs by H3K4me1. Our characterization of conserved players in repressive RNAPII transcription will inform efforts to characterize gene regulation through the act of pervasive lncRNA transcription in plants and beyond.    Quantification of hypocotyl length (mm) for 7 day, dark-grown wild type (n=30), qua1-1 (n=30), spt16-1 (n=30), ssrp1-2 (n=30), and double mutants qua1-1/spt16-1 (n=30) and qua1-1/ssrp1-2 (n=30). Three asterisks denote p<0.001 between qua1-1 and all other samples by Student's t-test. (C) Analysis of QUA1 transcripts in wild type, qua1-1, ssrp1-2, and the qua1-1/ssrp1-2 double mutant as well as in spt16-1 and the qua1-1/spt16-1 double mutant by northern blotting. Ethidium bromide (EtBr) staining of ribosomal RNA (rRNA) is used as a loading control.   (A) In wild type Arabidopsis RNAPII initiates transcription from promoter regions. Specific histone signatures such as H3 acetylation (grey circle) and H3K36me3 (yellow tri-hexagon) are associated with promoters, while RNAPII CTD Ser2 phosphorylation (green circle) and H3K36me2 (yellow di-hexagon) are associated with RNAPII elongation. (B) In qua1-1 and rfd1-1 mutants, upstream RNAPII transcription suppresses transcription initiation from downstream gene promoters (tandem Transcription Interference, tTI). The FACT complex contributes to tTI through chromatin remodeling activity. (C) Absence of FACT activity in rfd1-1 and qua1-1 mutants results in chromatin defects that trigger initiation of RNAPII transcription. H3K4me1 (turquoise) marks sites poised to function as intragenic TSSs in FACT mutants, presumably as H3K4me1 facilitates the transition to H3K4me3 that is linked to promoter activity. Intragenic RNAPII initiation in FACT mutants reduces tTI, as cryptic promoters may restore functional proteins. Supplementary Figure 4. qua1-1 suppression by spt16-1, but not sdg8-2. (A) The QUA1 and SSRP1 loci are linked on Arabidopsis chromosome 3. Genetic linkage prevents an analysis of qua1-1 suppression by ssrp1-2 using segregating populations. (B) Segregation analysis of qua1-1. We expect ¼ of the progeny from a qua1-1/QUA1 parent to be qua1-1/qua1-1 and hence stain by ruthenium red.

Acknowledgments:
We thank Jasmin Dilgen for technical assistance, Grégory Mouille and Henning Mühlenbeck for help with the ruthenium red staining assay, Jan Høstrup for plant care, Laura Brey for help in initial stages The authors declare that they have no conflict of interest.

Plant Growth
All Arabidopsis thaliana lines used in this study are in the Col-0 ecotype background, with the exception of the qua1-1 mutant, which is in WS ecotype background. Plants were grown in greenhouses or climate chambers with a 16h light/8h dark cycle at 22°C for general growth and seed harvesting. For seedlings grown on plates, the sterilized seeds were grown on 1/2 Murashige and Skoog (MS) medium containing 1% sucrose and supplemented with 0.5% Microagar. Information about T-DNA insertion mutant Arabidopsis lines is provided in Key Resource Table. Growing rfd1-1 plants For analysis of the homozygous rfd1-1 phenotype, seeds were sown in 96 well trays stratified for 2-3 days at 4°C. F2 analysis trays were grown in high light conditions (> 100 μE). White seedlings were counted 10 days later. To propagate rfd1-1 homozygotes, heterozygous rfd1-1 seeds were sterilized and sown on MS plates with phosphinotricin selection, covered in foil, and stratified for 2 days at 4°C. Seeds were light induced for 6-8 h in a growth chamber with light strength of 80-100 μE. Plates were covered in foil for 3 days, the plates were unwrapped and grown in low light (< 50 μE) for 3-4 weeks before transferring to soil. To isolate RNA, rfd1-1 homozygote seeds and corresponding wild type controls were sterilized and sown on MS plates as described above and grown in low light for two weeks. In order to collect enough material for ChIP, heterozygous rfd1-1 seeds were sterilized and sown on MS plates with phosphinotricin selection as described above and grown in low light for two weeks. Col-0 wild type controls were treated the same way, but without selection, for corresponding.

Ruthenium red staining
Seeds were sown in a 96 well plates containing 70 ul water. To synchronize germination, seeds were stratified at 4°C for 2-3 days. Germination of seeds was induced by light for 8-10 hours. The plates were wrapped in aluminium foil for 7 days. Etiolated seedlings were stained with 0.05% ruthenium red solution for 2 minutes. Seedlings were washed twice with water. Staining phenotype was recorded using a stereomicroscope.
Complementation constructs were generated using SMC330, a version of pEG302 (Earley et al., 2006) enabling Hygromycin selection following plant transformation. SMC330 was generated by replacing the phosphinotricin herbicide resistance gene with the Hygromycin resistance gene of pCambia1300. TIpQUA1:QUA1 and TIpRFD1:RFD1 were amplified from genomic wild type DNA using primers MLO727/728 and MLO414/442, respectively. The resulting PCR products were introduced into pENTR-D-Topo by topo cloning to generate entry vectors (SMC409 for TIpQUA1:QUA1 and SMC356 for TIpRFD1:RFD1). The entry vectors were used in a LR reaction with SMC330 to generate expression vector SMC410 (containing TIpQUA1:QUA1-FLAG construct) and SMC380 (containing TIpRFD1:RFD1-FLAG construct). The complementation constructs were then transformed into Agrobacterium tumefaciens strain GV3101 (pMP90) by electroporation under 2.5kV, 400Ω resistance and 25µF capacitance. The complementation assay was done using Agrobacterium-mediated transformation described in (Clough and Bent, 1998). Homozygous qua1-1 and heterozygous rfd1-1 Arabidopsis were used for complementation. Seeds from transformed Arabidopsis were screened for T-DNA integration by hygromycin resistance. Multiple independent single-locus insertions were identified by segregation analysis and tested for complementation and protein expression (Fig. 2).

GUS Staining and Fluorescence Imaging
The GUS staining assay was performed as previously described (Jefferson et al., 1987). X-Gluc substrate was vacuum infiltrated into Arabidopsis and N. benthamiana leaves. After staining, leaves were rinsed in 70% ethanol at room temperature until the chlorophyll was washed off. YFP fluorescence was quantified using BIO-RAD imager Gel Doc.

Western blotting
Equal amounts of plant material were harvested from plant tissue. Proteins were extracted in 2.5x extraction buffer (150 mM Tris-HCl pH 6.8; 5 % SDS; 25 % Glycerol; 0.025 % Bromphenolblue; 0.1 mM DTT). Proteins were separated by SDS-PAGE using precast 8-16 % Criterion TGX stain-free protein gels (Biorad) and transferred to PVDF membrane using a semi-dry Trans-blot Turbo transfer system (Biorad). Membranes were blocked (5% non-fat dried milk in PBS) for 1 hour at RT. anti-FLAG (Sigma F3165) was added overnight at 4°C with rotation. Membranes were washed with PBS before the addition of the anti-mouse HRP-conjugated secondary antibody (Dako P0161) for 1 hour at room temperature. Membranes were washed in PBST. Chemiluminescent signals were detected using Super-Signal West Pico Chemiluminescent (ThermoFisher) according to manufacturer's instructions.

Quantitative analysis was performed on captured DNA by qPCR (Biorad). See Supplementary Tables
S5 for oligonucleotide sequences. ChIP enrichments were calculated as the ratio of product of interest from IP sample to the corresponding input sample. For qua1-1 and corresponding wild type (ecotype WS), results were further normalized to an internal reference gene (ACT2). Error bars represent standard deviation resulting from at least three independent replicates.

RT-qPCR
RNA was isolated from 14 day old seedlings using Plant RNeasy Mini-Kits as per manufacturer's instructions (Qiagen). For RT-qPCR experiments, first strand complementary DNA synthesis was performed on Turbo DNase-treated (Ambion) RNA using oligo-dT primers and Superscript III (Invitrogen) as per manufacturer's instructions. Negative controls lacking the reverse transcriptase enzyme (-RT) were performed alongside all RT-qPCR experiments. Quantitative analysis was performed by qPCR (Biorad). See Supplementary Table S5 for oligonucleotide sequences. Data was normalized to an internal reference gene (ACT2). Levels in mutants represent relative expression compared to corresponding wild type.

Northern blotting
Northern analyses were essentially performed as previously described with only minor modifications (Marquardt et al., 2011). Briefly, 5 micrograms of total RNA were separated by electrophoresis on agarose-formaldehyde-MOPS gels and transferred to a nylon transfer membrane by capillary blotting in 10x SSC overnight. RNA was crosslinked to the nylon membrane by UV irradiation. Membranes were probed with single stranded cDNA probes generated by incorporation of radioactive α-32P-dTTP.
A Typhoon phosphoimager (GE Healthcare Life Sciences) was used for analysis.

TSS-seq Library Construction
TSSs were mapped genome-wide in Arabidopsis using 5'-CAP-sequencing as previously described (Pelechano et al., 2016). Briefly, 5 micrograms of DNase-treated total RNA were treated with CIP (NEB) to remove all non-capped species in the sample. Next, 5' caps were removed using Cap-Clip First-strand cDNA was generated using SuperScript III (Invitrogen) and random primers following manufacturer's instructions. Second-strand cDNA was generated using Phusion high-fidelity polymerase (NEB) and the BioNotI-P5-PET oligo as per manufacturer's instructions. Biotinylated PCR products were captured by streptavidin-coupled Dynabeads (Thermo Fisher Scientific), end repaired with End Repair Enzyme mix (NEB), A-tailed with Klenow fragment exo-(NEB), and ligated to barcoded Illumina compatible adapter using T4 DNA ligase (NEB). Libraries were amplified by PCR, size selected using AMPure XP beads (Beckman Coulter), pooled following quantification by bioanalyzer, and sequenced in single end mode on the following flowcell: NextSeq® 500/550 High Output Kit v2 (75 cycles) (Illumina).