The RNA helicase UPF1 associates with mRNAs co-transcriptionally and is required for the release of mRNAs from transcription sites

UPF1 is an RNA helicase that is required for efficient nonsense-mediated mRNA decay (NMD) in eukaryotes, and the predominant view is that UPF1 mainly operates on the 3’UTRs of mRNAs that are directed for NMD in the cytoplasm. Here we offer evidence, obtained from Drosophila, that UPF1 constantly moves between the nucleus and cytoplasm and that it has multiple functions in the nucleus. It is associated, genome-wide, with nascent RNAs at most of the active Pol II transcription sites and at some Pol III-transcribed genes, as demonstrated microscopically on the polytene chromosomes of salivary gland and by ChIP-seq analysis in S2 cells. Intron recognition seems to interfere with association and translocation of UPF1 on nascent pre-mRNA transcripts, and cells depleted of UPF1 show defects in several nuclear processes essential to correct gene expression – most strikingly, the release of mRNAs from transcription sites and mRNA export from the nucleus.

These observations are generally interpreted as evidence that UPF1 and related proteins are primarily required for nonsense-mediated mRNA decay (NMD), a conserved mRNA surveillance mechanism of eukaryotes that detects and destroys mRNAs at which translation terminates prematurely (Fatscher et al., 2015;He and Jacobson, 2015;Karousis et al., 2016;Kurosaki and Maquat, 2016). NMD is mainly regarded as a quality control mechanism that prevents cells from wastefully making truncated (and potentially toxic) proteins and that regulates the selective expression of specific mRNA isoforms during cell homeostasis and differentiation (Goetz and Wilkinson, 2017;Lykke-Andersen and Jensen, 2015).
Standard NMD models postulate that UPF1 monitors translation termination on ribosomes by interacting with a peptide release factor (eRF1 or eRF3). However, recent reports on mammalian translation systems have suggested, in contrast to earlier reports on other organisms (Czaplinski et al., 1998;Ivanov et al., 2008;Kashima et al., 2006;Keeling et al., 2004;Singh et al., 2008;Wang et al., 2001), that UPF1 does not bind to either of these. They suggested, instead, that UPF3B may contact release factors, slow the termination of translation and facilitate post-termination release of ribosomesand so fulfil the termination monitoring role that has been assigned to UPF1 (Gao and Wilkinson, 2017;Muhlemann and Karousis, 2017;Neu-Yilik et al., 2017). UPF1 is an ATP-driven helicase that unwinds RNA secondary structures and so can displace RNA-bound proteins (Bhattacharya et al., 2000;Chakrabarti et al., 2011;Czaplinski et al., 1995;Fiorini et al., 2015). Its helicase activity is required for NMD, but how this helps to target particular transcripts for NMD is not clear (Brogna et al., 2016;Brogna and Wen, 2009). UPF1 is predominantly associated with 3'UTRs of cytoplasmic mRNAs and it might be selectively recruited to or activated on NMD targets with abnormally long 3'UTRs (Karousis et al., 2016;Kurosaki and Maquat, 2016). However, UPF1 appears to bind mRNAs fairly indiscriminately, whatever the position of their stop codon or PTC and whether or not they include NMD-inducing features such as an abnormally long 3'UTR or an exon junction downstream of the stop codon (Hogg and Goff, 2010;Hurt et al., 2013;Zund et al., 2013).
UPF1 is most abundant in the cytoplasm and its roles discussed above depend on ribosomal translation and occur on cytoplasmic mRNAs. However, UPF1 traffics in and out of the nucleus, it interacts with chromatin, it co-purifies with the catalytic subunits of DNA polymerase δ, and UPF1 depletion impairs DNA replication and telomere maintenance (Ajamian et al., 2015;Azzalin and Lingner, 2006;Azzalin et al., 2007;Carastro et al., 2002;Chawla et al., 2011;Mendell et al., 2002). Moreover, there is evidence that UPF1 might contribute directly to RNA processing, at least in specific instances, and is required for nuclear export of HIV-1 genomic RNAs in HeLa cells (Ajamian et al., 2015;Brogna et al., 2016;de Turris et al., 2011;Flury et al., 2014;Varsally and Brogna, 2012).
In the present study we show direct evidence that UPF1 is globally involved in the formation and nuclear processing of mRNAs in Drosophila. First, we demonstrate that UPF1 is a highly mobile protein that constantly shuttles between the nucleus and cytoplasm, and its distribution in the cell, with more in the cytoplasm than the nucleus, approximately reflects that of mRNA. UPF1 associates with nascent transcripts on chromosomesmostly with Pol II transcripts, but also with some Pol III-transcribed genesand more of the transcript-associated UPF1 is bound with exons than with introns, suggesting that 5' splice sites might act as a roadblock to the 5'-to-3' transit of 4 UPF1 along the pre-mRNA. Most strikingly, UPF1 is needed for the efficient release of polyadenylated mRNA from most chromosomal transcription sites and for its export from nuclei. These observations show that UPF1 starts scanning pre-mRNA transcripts whilst they are still being assembled in ribonucleoprotein (RNP) complexes on chromosomes and suggest that it fulfils previously unrecognised role(s) in facilitating nuclear processes of gene expression and mRNA export. The broad and dynamic association with mRNAs redefines UPF1 from being primarily an NMD-inducing factor to being a global player in mRNA processing in the nucleus as well as in the cytoplasm, and might also explain why none of the prevailing models satisfactorily explains how UPF1 could target specific transcript to NMD.

Drosophila anti-UPF1 antibodies
To explore the functions of UPF1, we generated three monoclonal anti-peptide antibodies that target regions of Drosophila UPF1 outside the RNA helicase domain: one epitope in the N-terminal flanking regions (antibody 1C13 against Pep2), and two near the C-terminus (Ab 7D17 vs. Pep11; and Ab 7B12 vs. Pep12) (see Figure S1 and Supplementary Table S1). Each antibody detected UPF1 as a single band by Western blotting of Drosophila S2 cell extracts, with minimal cross-reactivity with other proteins, and also detected a second, larger band of the expected molecular mass in extracts from S2 cells that over-express UPF1-GFP (Figure 1A-1C; Figure S1). Unless otherwise indicated, antibody 7B12 was used in the experiments described below. As expected, UPF1 RNAi specifically reduced the amount of UPF1 in S2 cells ( Figure   1B) without affecting the levels of several other proteins we tested as controls ( Figure   1B).

UPF1 rapidly shuttles between nucleus and cytoplasm
We examined the subcellular localization of immunostained UPF1 in Drosophila salivary glands, which are made up of large secretory cells with polytene nuclei. UPF1 was most abundant in the cytoplasm and perinuclear region, and there was also distinct but less intense nuclear staining, mainly around the chromosomes ( Figure 1D).
Following cell fractionation of S2 cells, α-tubulin and RNA Pol II were, as expected, restricted to the cytoplasmic and nuclear fractions, respectivelyand a small proportion of the UPF1 co-purified with nuclei whilst most was in the cytoplasmic fraction ( Figure 1C).
Both cytoplasmic and nuclear UPF1 were also present in other larval tissues, with varying relative immunostaining intensities. Perinuclear and intra-nuclear UPF1 were more abundant in Malpighian tubules and gut ( Figure S2). In enterocytes (EC), staining was similar in the cytoplasm and within the nucleus, and the most intense UPF1 signal was perinuclear ( Figure S2B).
In salivary glands expressing UPF1-GFP ( Figure S3A) the perinuclear signals co-localised with binding of wheat germ agglutinin (WGA)a lectin that predominantly interacts with O-GlcNAc-modified nuclear pore proteins (Mizuguchi-Hata et al., 2013) and this UPF1-GFP may be associated with components of the nuclear pore complex, as has been proposed for S. cerevisiae UPF1 (Nazarenus et al., 2005). It is noteworthy, unexplained though, that UPF1 RNAi reduced the perinuclear WGA binding in salivary glands ( Figure 3SB); cells were also smaller, as would be predicted from its requirement in cell growth during Drosophila development (Metzstein and Krasnow, 2006).
Since UPF1 is present both in cytoplasm and nuclei, with the relative quantities varying between cell-types, we wondered how rapidly UPF1 shuttles between cell compartments. Such trafficking has been reported in HeLa cells, with UPF1 accumulating in the nuclei following treatment with leptomycin B (LMB) (Ajamian et al., 2015;Mendell et al., 2002); this drug selectively inhibits CRM1-mediated protein export from the nucleus in most eukaryotes (Fukuda et al., 1997).
We therefore explored the intracellular localization and dynamics of UPF1 in Drosophila salivary glands. Immunostained endogenous UPF1 and UPF1-GFP showed similar intracellular distributions, with an intense cytoplasmic signal and a weaker but still obvious signal in the regions occupied by chromosomes ( Figure S4A: the cytoplasmic texture of the salivary cells in these confocal images reflects the fact that they are packed with secretory vesicles at this stage of larval development). In glands treated with LMB for 60 minutes most of the UPF1-GFP was within the nucleus but largely excluded from the nucleolus ( Figure S4A, right panels), suggesting that UPF1 exit from the nucleus utilises a CRM1-dependent mechanism.
This UPF1 redistribution was rapid in living glands: UPF1 was accumulating in the nucleus by the earliest time we could collect images (within ~5-6 min), and much of the cell's UPF1 was in the nucleus within half an hour ( Figure S4B).
Heat-shock caused a similar redistribution of much of UPF1 from cytoplasm to nucleus, and this was partially reversed when the tissue was returned to its initial temperature ( Figure S4C). We next used two live cell imaging techniques -Fluorescence Loss in Photo-bleaching (FLIP) and Fluorescence Recovery after Photo-bleaching (FRAP) (Singh and Lakhotia, 2015) to examine the mobility of UPF1-GFP in salivary gland cells. FLIP revealed that sustained photobleaching of a small area of the cytoplasm led, within the continuously illuminated area, to an initial rapid decrease in UPF1-GFP fluorescence followed by a continued slower reduction. Fluorescence also declined steadily both elsewhere in the cytoplasm, and, more slowly, within the nucleus ( Figure   1E). These observations demonstrate ongoing diffusion of UPF1 throughout the cytoplasm, and that nuclear UPF1 can leave the nucleus and enter the photodepletable cytoplasmic UPF1 pool at a fairly steady rate.
The FRAP studies monitored the speed with which unbleached UPF1-GFP diffuses into and repopulates a photobleached region of the cytoplasm or nucleus. Almost all of the UPF1 in each cell compartment was rapidly mobile, and the halftime for repopulation of each bleached area was only a few seconds ( Figure 1F).
These observations indicate that UPF1 is freely mobile within cell compartments and that it constantly moves in and out of the nucleus by mechanisms that include the CRM1-dependent nuclear protein export pathway.

UPF1 associates with transcribing regions of the chromosomes
To gain insight into the role(s) of UPF1 in the nucleus, we used immunostaining to examine whether it associates with the polytene chromosomes of Drosophila salivary glands. These well-characterised giant interphase chromosomes are formed after multiple rounds of endoreplication without chromosomal segregation, and they provide a powerful system in which to visualise transcription and pre-mRNA processing at individual gene loci. UPF1 was present predominantly at interbands and puffs: cytologically distinct chromosome regions in which the chromatin is less condensed and that correspond to transcriptionally active sites ( Figure 2A). The immunofluorescence signal appears to be specific, as: a) UPF1-RNAi drastically depletes the endogenous UPF1 chromosomal signal (Figures S5A and S5B); and b) transgenically over-expressed UPF1-GFP, detected either by its fluorescence or with an anti-GFP antibody, shows a similar banding pattern at the chromosomes ( Figure S5C).
We then undertook double immunostaining of chromosomes for UPF1 and for Ser2 Pol IIthe form of Pol II that transcribes through the main body of genes which is characterised by having the C-terminal domain (CTD) of its largest subunit Ser2-phosphorylated (Boehm et al., 2003). Much of the UPF1 co-localized with Ser2 Pol II, as would be expected from this type of banding pattern ( Figure S6A).
The association of UPF1 with the chromosomes depends on transcription. This is illustrated by the changes in UPF1 immunostaining that followed heat-shock, which induces transcription at specific cytological puffs encoding heat-shock proteins and of hsrω lncRNAs at locus 93D (Lakhotia et al., 2012). This revealed a pattern of UPF1 association at heat shock puffs and of detachment from most other transcription sites ( Figure 2B). UPF1 was recruited to activated heat-shock genes that contained (33B, 63B, 64F, 67B, 70A and 93D) or lacked (87A, 87C and 95D) introns ( Figure 2B).
These observations suggested that UPF1 associates with genes that are being transcribed. UPF1 was also recruited to other genes following transcription activation, such as an ecdysone-inducible transgene (S136 at chromosomal position 63B) at normal temperature (Choudhury et al., 2016). No UPF1 was found at this locus while it is inactive, but when ecdysone activated the transgene it produced a cytologically distinguishable transcription puff with which UPF1 was associated ( Figure 2C).

UPF1 mainly associates with Pol II sites that are undergoing transcription and depends on the nascent transcript
We examined the association of UPF1 and of Ser2 Pol II with multiple gene loci by chromatin immunoprecipitation (ChIP) of S2 cell extracts, followed by high-throughput DNA sequencing (ChIP-Seq). UPF1 was associated with many transcriptionally active genes, most of which are Pol II transcription sites. Figure 3A shows enrichment profiles of UPF1 and of Ser2 Pol II across a representative chromosome region. Actin5C provided a striking example of correspondence between the ChIP-seq and polytene immunostaining results: it was one of the most UPF1-enriched genes in the ChIP-seq data (Table S2, Figure S9 shows the UPF1 ChIP-seq profile of Actin5C) and displayed one of the brightest UPF1 chromosomal signals at the gene locus corresponding to interband 5C on the X chromosome ( Figure   2A). The ChIP-seq data also show UPF1 association with a few Pol III genes (Table   S2, to be discussed later).
The enrichment profile of UPF1 at Pol II loci closely followed that of Ser2 Pol II, and UPF1 enrichment was greatest at highly expressed genes ( Figure 3A; and Figure S8A and S9 show additional examples of UPF1-enriched genes). There was a close correlation between UPF1 and Ser2 Pol II ChIP-seq signals, and also between these and mRNA levels ( Figure 3B and 3C). Real-time PCR was used to validate the ChIP-seq data at several genes, both in S2 cells and salivary glands ( Figure S7; and other examples are shown below). UPF1-RNAi drastically reduced the UPF1 enrichment at transcription sites, both confirming the specificity of the antibody and validating the ChIP protocol ( Fig S7C).
A metagene analysis of the ChIP-seq data shows that UPF1 is associated with genes, and particularly with highly expressed genes (blue trace), throughout their transcription units ( Figure 3D) whereas Ser2 Pol II typically shows higher loading around transcription start sites (TSS)as previously reported in Drosophila and other organisms (Adelman and Lis, 2012;Muse et al., 2007). Typically, therefore, most gene-associated UPF1 was further downstream than the TSS-proximal Ser2 Pol II peak, especially at highly expressed genes ( Figure 3E: striking examples of this pattern are the Su(z)2 and Psc genes ( Figure 3A) and the α-Tub84B gene ( Figure S8A).
A comparison of the UPF1 loading of genes with different Ser2 Pol II loading profiles suggests that UPF1 association depends on transcription elongation: UPF1 did not associate with genes at which Ser2 Pol II was associated only with the TSS pausing site and which were not being actively transcribed (e.g. Adam TS-A, panel 5 in Figure  S8A).
The association of UPF1 with Pol II transcription sites is partially sensitive to RNase treatment, suggesting that UPF1 loads onto nascent RNA. This was apparent both for immunostained UPF1 on polytene chromosomes ( Figure S6B and C) and when assayed by ChIP at specific genes by qPCR in S2 cells ( Figure S7D). UPF1 association was, though, less sensitive to RNase treatment than that of the RNA binding protein hnRNPA1 ( Figure S6B-C), which is almost all detached following the same RNase treatment. Some of UPF1 co-purifies with Ser2 Pol II in a standard immunoprecipitation of S2 cell nuclear extracts, the interaction is similarly sensitive to RNase treatment though ( Figure S7E): less than that of hnRNPA1, but comparable to that of eIF4AIII, one of the exon junction complex (EJC) proteins that are loaded onto nascent RNAs (Choudhury et al., 2016).
We also examined the effect on salivary glands of 5,6-dichloro-1-beta-D-ribofuranosylbenzimidazole (DRB), a drug that blocks Pol II transcription by inhibiting Ser2 phosphorylation (Bensaude, 2011). In the presence of DRB, unphosphorylated Pol II (Pol II) initiates transcription but does not engage in productive elongation as this would require Ser2-phosphorylated Pol II (Ser2 Pol II) (Adelman and Lis, 2012). DRB treatment left interbands and puffs cytologically unaffected, as expected, but it markedly reduced the amount of UPF1 associated with gene loci ( Figure S6D-E), providing further evidence that transcript elongation into the body of the gene is needed for this association to occur. DRB also reduced the association of UPF1 and Ser2 Pol II with genes, such as the highly expressed RpL23A, in S2 cells ( Figure 3F).

UPF1 at Pol III transcription sites
UPF1 was found mainly at Pol II transcription sites, most of which are protein-coding genes, but our ChIP-seq data also revealed it at a minority of Pol III genes. The latter included 7SK and both paralogous genes of 7SL snRNAs ( Figure S8B)but not, for example, the much more numerous Pol III-transcribed tRNA genes ( Figure S8C, Table   S2).

Intron recognition interferes with UPF1 association with nascent transcripts
UPF1 was recruited both to intron-containing and intronless genes that were undergoing transcription ( Figure S9A, and see also the earlier discussion of heat-shock gene activation), so recruitment did not depend on pre-mRNA splicing.
Within intron-containing genes, however, more UPF1 was associated with exons than with intronsas can be seen in the ChIP-seq profiles of highly UPF1-enriched genes such as Xrp1 ( Figure 4A; and Figure S9 shows other examples of genes displaying this pattern).
This exon-biased UPF1 enrichment was confirmed by real time PCR in multiple ChIP experiments ( Figure 4B); and it is genome-wide, as demonstrated by comparing UPF1 association with introns and with their flanking exons in the ChIP-seq data from many genes ( Figure 4C, UPF1 enrichment is significantly higher for both the left (P = 6.737e-8) and the right flanking exon (2.391e-9); for details of how we corrected for possible bias in chromatin fragmentation or sequencing coverage, see Methods). This The lower frequency with which UPF1 associated with introns suggested that some features of unspliced transcripts must interfere with the UPF1 interaction. We hypothesised that 5' splice sites (5'ss) at the starts of introns, where the initial U1 snRNP spliceosome complex would bind, might act as road-blocks to UPF1 translocation along nascent pre-mRNAs and that removal of U1 might allow UPF1 to move on through the intron ( Figure 4G). We therefore used ChIP in S2 cells to compare the enrichment of UPF1 at exons and introns in the Xrp1 gene in cells that had been depleted either of the U1 snRNP protein U1-70K or of Y14 or eIF4III (two of the EJC proteins that bind the nascent pre-mRNA but are not likely to play a direct splicing role in Drosophila; see (Choudhury et al., 2016)). The normal bias towards UPF1-exon association in Xrp1 transcripts was abolished in the U1-70K-depleted cells but persisted in cells depleted of eIF4AIII or Y14 ( Figure 4F).
Moreover, genes with the most marked exon-biased UPF1 enrichment, such as Xrp1, are efficiently co-transcriptionally spliced (see the Nascent RNA-seq profile in Figure   4A), whereas genes with no detectable exon-biased UPF1 enrichment, such as CG5059, are poorly co-transcriptionally spliced ( Figure S9C) and are typically expressed at low levels, as reported (Khodor et al., 2011). It seems therefore, that intron recognition interferes with the association of UPF1 with the unspliced nascent transcript.

Pol II tends to stall near Transcription Start Sites (TSSs) in UPF1-depleted cells
We next determined whether the availability of UPF1 influences the relative amounts of unphosphorylated Pol II and Ser2 Pol II that associate with genes in S2 cells.
Both unphosphorylated and, to a lesser extent, Ser2-phosphorylated Pol II were most frequently associated with the start of genes ( Figure 5A-5D). Its loading peaked 20 to 60 nucleotides downstream of the TSS (as shown in the expanded depiction in Figure   5B that is based on many more genes), at a position that corresponds to the average distance from TSSs to Pol II pausing sites (Adelman and Lis, 2012). Significantly more unphosphorylated Pol II accumulated there in cells depleted of UPF1 (P = 0.011; light blue line in Figure 5B, based on quantifying the aggregate Pol II signal over a +/-100bp span at each of 25440 TSSs): many TSSs showed a 1.2-fold or greater increase in the UPF1-depleted cells (Table S3). Conversely, the amount of Ser2 Pol II associated with these TSSs was unchanged or marginally reduced in UPF1-depleted cells ( Figure 5C-5D).
The increase in unphosphorylated Pol II loading downstream of the TSS in UPF1-depleted cells, alongside fairly constant Ser2 Pol II loading, is illustrated here by Xrp1 and RpS3A ( Figure 5E), two genes that are highly transcribed and show strong UPF1 association ( Figure 4A, Table S2).
We also assessed transcription by monitoring the repopulation of genes by newly phosphorylated Ser2 Pol II following the withdrawal of DRB treatment. As expected, DRB treatment of S2 cells led to a drastic depletion of Ser2 Pol II and accumulation of unphosphorylated Pol II in cell extracts ( Figure 5F, compare Control lanes 9 and 10 with DRB-treated lanes 1 and 2). Ser2 Pol II levels began to recover soon after DRB removal, and were similar to those of untreated control cells within 10 minutes ( Figure   5F, compare lanes 7-8 with control lanes 9-10). However, this recovery seemed slower in the UPF1-depleted cells ( Figure 5F, compare lanes 3 vs. 4 and 5 vs. 6). A similarly blunted recovery of gene-associated Ser2 Pol II in UPF1-depleted cells was detected by ChIP at the two gene loci (Socs36E and Xrp1) that were assayed by real-time PCR ( Figure 5G).
Comparable genome-wide observations were made using polytene chromosome spreads. There were no obvious changes in Ser2 Pol II distribution when UPF1 was simply depleted ( Figure S10A). When, though, UPF1 was depleted and the glands were also DRB treated, there was then a delay in the recovery of the Ser2 Pol II signal when DRB was removed (Figures S10B, S10C).
Cumulatively, these observations suggest that UPF1 might, by associating with nascent transcripts, influence the phosphorylation of Pol II and hence the transcription of some genes.

UPF1 depletion leads to nuclear mRNA retention
We also assessed whether depleting UPF1 in the salivary gland cells of 3 rd instar larvae would have any effect on mRNA release from transcription sites and its subsequent processing and export from the nucleus.
First we examined the overall cellular distribution of poly(A) RNAwhich is referred to from here on simply as poly(A)by oligo(dT) FISH (fluorescence in situ hybridization): this should detect mRNA that has been transcribed, spliced, released from Pol II and polyadenylated. In wild-type cells poly(A) was abundant and fairly evenly distributed throughout the cytoplasm, as would be expected for mature mRNA, and there was little in the nuclei ( Figure 6A). By contrast, the nuclei of UPF1-depleted cells retained a substantial amount of poly(A), and the cells appeared to contain less cytoplasmic poly(A) than wild-type cells. Much of the nuclear-retained poly(A) in the UPF1-depleted cells formed large cluster(s) in inter-chromosomal spaces ( Figure 6A) that seemed neither to be linked to or in the proximity of any specific chromosomal region(s) or defined transcription site(s).
An appreciable amount of poly(A) signal, which was not within clusters, was clearly at the chromosomes though, in the UPF1-depleted cells ( Figure 6A, panels III and VI).
We therefore used oligo(dT) FISH on polytene chromosome spreads to compare wild-type and UPF1-depleted cells and to assess whether there is retention of poly(A) near transcription sites. There was little poly(A) associated with most of the wild-type chromosomes. However, a few interbandssuch as 2C at the distal end of the X chromosome ( Figure 6B)showed clear poly(A) signals ( Figure 6B, left panel), suggesting that some completed mRNAs that have been cleaved and polyadenylated remain associated, at least briefly, with transcription sites. Additionally, since UPF1 was obviously not associated with 2C (see Figure 2A), the poly(A) accumulation at 2C in wild-type cells may be a consequence of UPF1 not being normally associated with this transcription site.
Both the number of transcriptional sites showing poly(A) accumulation and the amount of poly(A)RNA associated with these sites were strikingly increased in UPF1-depleted cells ( Figure 6B, right panel). For example, there was no visible poly(A) accumulation at site 5C, which corresponds to the highly transcribed Actin5C gene, in wild-type, but this band was obviously fluorescent in UPF1-depleted cells.
Another example was site 2Bwhere constitutively expressed sta and rush are probably the most active genes at this larval stagewhich showed a faint poly(A) signal in wild-type glands and a strong signal in UPF1-depleted cells. UPF1 was clearly associated with these transcription sites (2B and 5C) on polytene chromosomes (Fig 2A) and in S2 cells (as detected by ChIP: see Table S2 and Figure   S9 for the UPF1 profile of Actin5C).
Cumulatively, these data make it clear that UPF1 plays important role(s) both in the release of mRNAs from transcription sites and in their transport out of the nucleus ( Figure 6C shows a cartoon of a transcription site of either a wild-type or UPF1 depleted cell with or without mRNA retention).

Discussion
The RNA helicase UPF1 is usually most abundant in the cytoplasm and is mainly discussed in relation to NMD, leading to the common assumption that it acts mainly on mRNPs that have been exported from the nucleus. In contrast, we present evidence that UPF1 moves constantly within and between cell compartments, that it interacts with mRNA both in the nucleus and the cytoplasm, and that it starts its mRNA association(s) at transcription sites, cotranscriptionally and before pre-mRNA processing is complete.
Within the nucleus we found UPF1 associated with many actively transcribing Pol II sites, to which it seems mainly to be recruited by an interaction with nascent pre-mRNA. More of this transcript-tethered UPF1 is associated with exons than with the introns that separate them. However, this distinction is lost in cells depleted of the spliceosome component U1 snRNP, suggesting that when U1 snRNP is bound to the 5'ss of an intron at the initial stage of splicing, it may hinder UPF1 translocation along the pre-mRNA and cause it to dissociate. It is conceivable that this scanning of pre-mRNAs by UPF1 might influence splice site recognition and pre-mRNA splicing in a manner consistent with reported observations that UPF1 depletion provokes changes in the relative concentrations of many alternatively spliced transcripts in S2 cells (Brooks et al., 2015) and with the model offered in Figure 4G. The association of UPF1 with nascent transcripts seems to be dynamic, and its putative 5'-to-3' scanning along RNA seems likely to be fast and, at least on intron-containing pre-mRNA, discontinuous. This pattern also suggests that when it encounters a steric block that cannot be removed UPF1 must be capable of quickly dissociating and re-loading elsewhere on the transcript. In vitro, UPF1 can translocate along RNAs over long distancesbut only at a maximum scanning velocity of ~80 base/min (Fiorini et al., 2015), which is much slower than the 2-3 kb/min of Pol II (Fiorini et al., 2015;Fukaya et al., 2017); possibly UPF1 translocates faster in vivo.
We did not detect any major impairment of Pol II transcription in UPF1-depleted cells, but Pol II pausing downstream of the TSS was more apparent at some genesfor example, at Xrp1, a strikingly UPF1-associated gene. And transcription appears to recover more slowly from DRB inhibition in UPF1-depleted cells.
The most striking effects of UPF1 depletion were retention of poly(A) RNA at transcription sites and then its failure to be exported effectively from the nucleus.
Completed mRNA transcripts that have been cleaved and polyadenylated are normally expected to be speedily released from transcription sites, but our data show that this is not always the case. We have both: a) identified some sites on polytene chromosomes that apparently accumulate poly(A) RNA even in wild-type glands; and b) shown that most of the active Pol II genes accumulate poly(A) RNA in UPF1-depleted glands.
Poly(A) RNA accumulation in UPF1-depleted cells is most marked at genes with which UPF1 associates strongly in wild-type, such as Actin5C (as shown both microscopically and by ChIP-seq). Conversely, those few transcription sites at which poly(A) accumulates even in wild-type cells, may not normally be associated with UPF1, a striking example is transcription site 2C on the polytene chromosomes, which showed the most apparent poly(A) accumulation but no obvious UPF1 association.
Evidence of retention of poly(A) and specific mRNAs in discrete nuclear foci or "dots" has previously been reported in cells defective in RNA processing, initially in mRNA export and processing mutants in yeast (Jensen et al., 2001). But whether these foci corresponded to intranuclear mRNP aggregates at sites adjacent to rather than at transcription sites is not clearand so far these nuclear poly(A) foci have only been reported in cells in which one of several RNA processing reactions are impaired (Abruzzi et al., 2006;Paul and Montpetit, 2016). Whether the previously described "dots" correspond to the poly(A) clusters that accumulate in the inter-chromosomal spaces of UPF1-depleted nuclei and/or to accumulations of poly(A) at transcription sites, which we identified here, remains to be determined.
In summary, our results indicate that UPF1 plays an important genome-wide role in nuclear processes of mRNA formation and in their release from transcription sites and export to the cytoplasm, at least in Drosophila. Possibly, in the absence of UPF1 function mRNPs acquire or remain in native conformations that hinder their release from the chromosome and make them prone to aggregation and hence nuclear retention. This global role could explain better than NMD why UPF1 is universally conserved in eukaryotes and why its depletion noticeably affects the expression of a large fraction of the genome.

Drosophila Stocks
Flies were reared in standard corn meal fly food media at 24°C. The yw strain was used as wild type. UAS-UPF1-RNAi (43144) and UAS-GFP-UPF1 (24623) were obtained from the Bloomington stock centre. The forkhead (Fkh) Gal4 has a salivary gland specific expression from early stage of development (Henderson and Andrew, 2000).
The transgenes expressing the lacO-tagged and ecdysone inducible S136 construct was described before (Choudhury et al., 2016).

S2 cells were cultured in Insect-XPRESS media (Lonza) supplemented with 10%
Fetal Bovine Serum (FBS) and 1% Penicillin-Streptomycin-Glutamine mix (P/S/G, Invitrogen) at 27°C. To make the RNAi constructs for UPF1, eIF4AIII, Y14 and snRNPU1-70k mRNA, the specific sequences were PCR amplified from S2 cell genomic DNA by using corresponding primer pairs (Table S4). Along with the desired gene sequence, all these primer pairs carried the T7 promoter sequence (in bold) at their 5' end (5'-TTAATACGACTCACTATAGGGGAGA-3'). The amplified PCR fragments were purified using Monarch® PCR and DNA Cleanup Kit (T1030S, NEB) and dsRNA was synthesized using the T7 RiboMAX express RNAi system (P1700, Promega). To induce RNAi, a six-well culture dish was seeded with 10 6 cells/well in serum-free media and mixed with 15 µg of dsRNA/well. Following 1 hr incubation at RT, 2 mL of complete media was added to each well and the cells were incubated for the next three days to knockdown the corresponding RNA and then harvested. The RNAi efficiency of UPF1, eIF4AIII and Y14 was measured by Western blotting while snRNPU1-70k was measured by real time PCR.

Generation of monoclonal antibodies against Drosophila UPF1
Antigens design, preparation, mice immunization and hybridoma generation were carried out by Abmart (Shanghai). Twelve peptide sequences predicted to be highly immunogenic were selected from D. melanogaster UPF1 (Table 1) and cloned in-frame in an expression vector to produce a recombinant protein incorporating all 12 antigens which were used as the immonogen (Abmart, SEAL TM technology). Hybridoma clones were generated and used to induce 18 ascites, which were then screened by Western blotting of S2 cell protein extracts. Out of these, three that showed a single band of the expected size and minimal cross-reactivity were selected and more of the monoclonal antibodies subsequently purified from the corresponding hybridoma cell culture in vitro.
Unless otherwise specified, 7B12 was used as the anti-UPF1 antibody throughout this study.

Larval tissue immunostaining
Whole-mount immunostaining was performed as previously described (Choudhury et al., 2016). In brief, the internal organs of 3 rd instar larvae were dissected in 1X PBS (13 mM NaCl, 0.7 mM Na2HPO4, 0.3 mM NaH2PO4, pH 7.4) and fixed in 4% formaldehyde for 20 minutes at RT. Tissues were washed in 1XPBS followed by 1% Triton X-100 treatment for 20 minutes. Tissues were washed and incubated in blocking solution (10% Fetal Bovine Serum (FBS), 0.05% Sodium Azide in 1X PBS) for 2 hrs at RT and then incubated in primary antibodies at 4 0 C overnight. Tissues were washed and further incubated with appropriate fluorescent-tagged secondary antibodies for 2 hrs typically. After washing, tissues were incubated in DAPI (4-6-diamidino-2-phenylindole, Sigma-Aldrich, 1 μg/mL) for 10 minutes and mounted in PromoFluor Antifade Reagent (PK-PF-AFR1, PromoKine) mounting medium and examined using a Leica TCS SP2-AOBS confocal microscope.

LMB, DRB and larvae heat shock treatment
Wandering 3  125µM) for 1 hr at RT. For heat shock response, larvae were placed in a pre-warmed microfuge tube lined with moist tissue paper and incubated in water-bath maintained at 37+1 0 C for 1 hr.

Live cell imaging (FRAP and FLIP)
Fluorescence recovery after photobleaching (FRAP) and fluorescence loss in photobleaching (FLIP) methods have been previously described (Klonis et al., 2002).
Salivary glands expressing UPF1-GFP were dissected from 3 rd instar larvae and mounted as a hanging drop in M3 media. For the FRAP the region of interest (ROI, a circle of fixed diameter) was rapidly photobleached with 100 iterations of 100% power Argon laser (488 nm) exposure. Subsequent recovery of fluorescence in the photobleached region was examined at defined time intervals. As a control, fixed cells were examined to confirm irreversible photobleaching. FRAP experiments were carried out on salivary glands at room temperature. The fluorescence signal in ROI was normalized and data analysed following published methods (Phair and Misteli, 2000;Singh and Lakhotia, 2015). FLIP experiments were done as previously described (Phair and Misteli, 2000). Following an acquisition of five control images, GFP fluorescence in ROI1 was continuously photobleached with Argon laser (488 nm) at 100 % power by 50 iterations. The loss in fluorescence in another region of interest, the ROI2 was measured for the same length of time. Fluorescence intensities at ROI1 and ROI2 were normalized and data analysed as described (Nissim-Rafinia and Meshorer, 2011). Both photobleaching experiments have been done using a Leica TCS SP2-AOBS confocal microscope.

Polytene Chromosomes Immunostaining
Apart from the changes detailed below, the procedure was mostly as previously

RNA-seq
Extracted RNA samples were quantified using a Nanodrop-8000 Spectrophotometer (ThermoFisher ND-8000-GL) to assess quality and to determine concentrations.
Aliquots of each sample were diluted to ~5ng/µl, and tested with an Agilent Multiplexed libraries were sequenced (50-bp single-end reads) on a HiSeq4000 sequencer.

CHIP-seq and RNA-seq data analysis
ChIP-seq and RNA-seq data were initially viewed and analysed using the Lasergene  (151623) Table S4).  Table S1). The peptides indicated by red color produced the monoclonal antibodies with highest specificity, as shown below ( 7B12 is shown in Figure 1A.