Hypoxia regulates overall mRNA homeostasis by inducing Met1-linked linear ubiquitination of AGO2 in cancer cells

Hypoxia is the most prominent feature in human solid tumors and induces activation of hypoxia-inducible factors and their downstream genes to promote cancer progression. However, whether and how hypoxia regulates overall mRNA homeostasis is unclear. Here we show that hypoxia inhibits global-mRNA decay in cancer cells. Mechanistically, hypoxia induces the interaction of AGO2 with LUBAC, the linear ubiquitin chain assembly complex, which co-localizes with miRNA-induced silencing complex and in turn catalyzes AGO2 occurring Met1-linked linear ubiquitination (M1-Ubi). A series of biochemical experiments reveal that M1-Ubi of AGO2 restrains miRNA-mediated gene silencing. Moreover, combination analyses of the AGO2-associated mRNA transcriptome by RIP-Seq and the mRNA transcriptome by RNA-Seq confirm that AGO2 M1-Ubi interferes miRNA-targeted mRNA recruiting to AGO2, and thereby facilitates accumulation of global mRNAs. By this mechanism, short-term hypoxia may protect overall mRNAs and enhances stress tolerance, whereas long-term hypoxia in tumor cells results in seriously changing the entire gene expression profile to drive cell malignant evolution.

P ost-transcriptional regulation of mRNA through mRNA deadenylation, translational repression and mRNA destabilization by miRNA-induced silencing complexes (miR-ISC) is an essential mechanism in modulating mRNA turnover 1 . AGO2, a key effector of miRISC, assembles with miRNAs to form the basic core of miRISC, and simultaneously recruits a flexible scaffold protein TNRC6/GW182 to make a bridge with the multiple downstream effectors, which prompt posttranscriptional silencing of complementary targeted mRNA 2 . In the case of more perfectly complementary base, miRNA-targeted mRNAs can be directly cleaved by AGO2 through its nuclease activity 3 . According to computational predictions and miRNAtargeted genome-wide identifications, more than half of proteincoding genes are quantitatively supervised by the miRNA regulation system 4 . Dysregulation of miRNA biogenesis and disruption of the miRNA pathway are found in human cancer samples, which provides the compelling evidence that miRNAs are crucially important for a wide variety of human diseases, including cancers 5,6 . Dissecting definitely mechanisms of miRNA biogenesis and miRNA-mediated gene silencing, especially in response to physiological or microenvironmental stress stimuli 7 , provides a promising therapeutic strategy for cancers.
Clinical and experimental evidence indicate that hypoxia is an important microenvironmental factor that promotes tumorigenesis and cancer development. Multiple molecular mechanisms underlying hypoxia promoting tumor evolution are previously described, including its inducing hypoxia-inducible factors (HIFs) and other transcription factors to activate gene transcriptions, enhancing specific oncogenic mRNA translation, as well as affecting the protein stability 8,9 . In addition, hypoxia also leads to inactivation of the demethylases KDM5A and KDM6A, which increase the levels of H3K4me3 and H3K27me3, respectively, and regulate the transcription of downstream genes 10 . However, whether hypoxia is involved in the regulation of the mRNA homeostasis has received little attention, although it is reported that hypoxia stabilizes some specific mRNAs by inhibiting the mechanism of nonsense-mediated RNA decay (NMD) 11 . In cancer cells, whether under hypoxia overall mRNA homeostasis is regulated in a NMD independent manner is not explored. In response to hypoxia, during tumorigenesis miRNA biogenesis is modulated through post-translational modifications of core processing enzymes of the miRNA pathway [12][13][14] , but whether hypoxia possesses the capacity in modulating miRNA functional efficiency and its underlying molecular mechanism is still unexplored.
Protein ubiquitination system has been emerged as a protein quantity detective and a huge cellular signaling transduction net for controlling signaling pathways 15 . The linear ubiquitin chain assembly complex (LUBAC), consisting of HOIP (RNF31), HOIL-1L (heme-oxidized IRP2 ubiquitin ligase 1 L) and SHAR-PIN (Shank-associated RH domain-interacting protein), has been identified as the unique ubiquitin ligase that specifically catalyzes Met 1 -linked linear polyubiquitin chain to its substrates, in which the N-terminal methionine (Met) of an ubiquitin can form a peptide bond with the C-terminal glycine (Gly) of another ubiquitin [16][17][18][19] . On the contrary, the Met 1 -linked linear polyubiquitination (termed as M1-Ubi) is negatively regulated by OTULIN (Gumby/FAM105B), CYLD and A20 with different mechanisms [20][21][22][23] . M1-Ubi endows with an essential role in activating inner inflammation and immune signaling [17][18][19] . HOIP or HOIL-1L deficiency in mice results in embryonic lethality at mid-gestation 24,25 and SHARPIN deficiency in mice results in immunodeficiency and severe multiple-organs inflammation disease 26 , respectively. Native deficiency of either HOIP or HOIL-1L in human causes autoimmunity and inflammatory disease 27,28 . Meanwhile, native mutations of HOIP in human is closely related to lymphoma 29 . Since hundreds of linear polyubiquitin-modified substrates are identified by through mass spectrometry analysis combined with lysine-less internally tagged ubiquitin strategy 30 , M1-Ubi may be involved in variously cellular processes. Thus far, cellular signaling regulated by M1-Ubi in non-inflammation and non-immune signaling are still mostly unknown.
Here we find that hypoxia suppresses miRNA-mediated gene silencing, and the interactions of miRNA-targeted mRNAs with AGO2 are extensively inhibited under hypoxia stress. Intriguingly, we demonstrate that hypoxia induces the recruitment of LUBAC components HOIP and HOIL-1L to AGO2, thereby catalyzing M1-Ubi of AGO2. The M1-Ubi of AGO2 blocks its association with targeted mRNAs and subsequently decreases miRNA-mediated gene silencing, thus resulting in the accumulation of miRNA-targeted mRNAs. Thus, our studies reveal that M1-Ubi of AGO2 is an essential mechanism modulating mRNA turnover in the miRNA pathway in hypoxic cancer cells.

Results
Hypoxia impairs miRNA-targeted mRNA loading to AGO2. To investigate the role of hypoxia in modulating mRNAs loading to AGO2, RNA immunoprecipitation sequencing (RIP-Seq) of AGO2 were performed by using stable HeLa cells expressing Flag-AGO2, which were separately treated with normoxia (21% O 2 ) and hypoxia (1% O 2 ) for 24 h. 11,084 and 13,176 of mRNA transcripts associated with AGO2 were identified under normoxia and hypoxia conditions, respectively, and there were 7638 mRNA transcripts overlapping between normoxia (68.9%) and hypoxia (58.0%) conditions ( Supplementary Fig. 1a, Supplementary Data 1). Cumulative fraction analyses showed that hypoxia significantly decreased the interactions of mRNA transcripts with AGO2 (P = 2.272E−10, Mann-Whitney U test) (Fig. 1a) and distribution plots further showed that hypoxia attenuated 2100 mRNA transcripts and increased 1528 mRNA transcripts associated with AGO2 (based 2 fold change), respectively (Supplementary Fig. 1b). These results indicate that hypoxia possesses an unrevealed role in suppressing mRNAs loading to AGO2.
MiRNA-targeted mRNAs recruiting to AGO2 for translation repression or decay is mediated through miRISC 2 . To test whether the decreased mRNAs loading to AGO2 is due to the change of miRNA profile under hypoxia, the high-throughput sequencing for miRNA (miRNA-Seq) was conducted, showing that the biogenesis of a small fraction of miRNAs was influenced under hypoxia ( Supplementary Fig. 1c, Supplementary Data 2). Similar results were obtained by using other groups' data of miRNA-seq and microarray in MCF7 31 , miRNA-Seq in 786-O 32 and murine C166 endothelial 33 Fig. 1d-f). Core catalytic enzymes of the miRNA pathway and some of miRNAs which are even reported regulated by hypoxia [12][13][14] , were determined to show that the expression levels of these proteins (DICER, Drosha, AGO2 and TARBP2) were inconsistently and slightly changed in 7 different cell lines under hypoxia ( Supplementary Fig. 1g). In response to hypoxia, EGFR interacts and phosphorylates AGO2 at the position of Tyr393, which inhibits the interaction of AGO2 with DICER to modulate parts of long-loop structure pre-miRNAs processing 12 , such as pre-miR-192, pre-miR-31 and pre-miR-193a, but not of the shortloop structure pre-miR-21. Consistent with this, the expression of let-7a, miR-21 and miR-19b were not changed under hypoxic ( Supplementary Fig. 1h). These results illustrate that hypoxia stress has a prior effect on the biogenesis of only some special miRNAs, but which is dependent on cell types.
To further verify this, we analyzed the associations between AGO2 and mRNAs targeted by the top 10 non-difference miRNAs ( Supplementary Fig. 2g) under hypoxia condition.
Considering that one mRNA might be targeted by several miRNAs, we chose those mRNAs, which are targeted only by top 10 non-difference miRNAs but not by those of hypoxia-reduced miRNAs, for analysis of the RIP-Seq data, and showed that hypoxia significantly attenuated the association between AGO2 and targets with sites of only top 10 non-difference miRNAs (Fig. 1d). Notably, the enriched abundance of mRNA transcript peaks with RIP-Seq data showed that hypoxia attenuated the interaction of MYC and HMGA2 mRNA transcripts with AGO2 ( Fig. 1e, f). These results demonstrate that hypoxia decreases miRNA-targeted mRNA loading to AGO2.
Hypoxia inhibits miRNA-targeted mRNA turnover. To answer whether the suppressed association of miRNA-targeted mRNAs with AGO2 by hypoxia inhibits mRNA decay, the highthroughput sequencing of mRNA (RNA-Seq) were conducted (Supplementary Data 3). Transcript mRNAs FPKM (Fragments per kilobase of exon per million fragments) >1 presented in both samples were analyzed in parallel. The expressions of 1578 targets by the top10 non-difference miRNAs and 1413 non-miRNA targets were identified (mean FPKM > =30 is considered as the effective candidates) and shown as distribution plots (Supplementary Fig. 2h). The results showed that hypoxia significantly increased the abundance of the top 10 non-difference miRNA targets than that of non-miRNA targets (Fig. 1g). These top 10 non-difference miRNA targets were further subdivided into as 6mer, 7mer and 8mer miRNA-seed complementarily paired targets, showing that hypoxia enriched the abundance of miRNA targets with a single 8mer site more than those of miRNA targets with a single 7mer and 6mer site (Fig. 1h). In addition, we employed a let-7 miRISC luciferase reporter assay, in which fourrepeated let-7a binding site (BS) sequences were reconstructed into the 3′-UTR of Renilla luciferase in the psiCHECK2 vector getting a reporter construct psiCHECK2-4xlet-7a-BS, and the miRISC luciferase activity was normalized by Firefly luciferase (Supplementary Fig. 2i). We transfected this reporter into 293T cells, and then treated with hypoxia or CoCl 2 for indicated times, the results showed that hypoxia stress significantly increased the inhibition of the luciferase reporter activities (Fig. 1i). Furthermore, we transfected let-7a miRNA mimics and negative control into 293T cells stably expressing GFP-4xlet-7a-BS for 24 h, and subsequently treated with hypoxia for 24 h. The GFP protein expression detection by Western blot showed that hypoxia recovered the GFP mRNA decay by let-7a mimics ( Supplementary Fig. 2j). Meanwhile, a similar experiment of siRNA targeted to endogenous PTEN mRNA in HeLa cells was performed, and the results showed that hypoxia similarly recovered the endogenous PTEN decay by siRNA ( Supplementary  Fig. 2k). Moreover, gene ontology (GO) analysis revealed that several distinct gene clusters such as cell division, DNA repair and mRNA processing were gathered under hypoxia ( Supplementary  Fig. 2l). Taken together, all the above results illustrate that hypoxia impairs miRNA-targeted mRNA loading to AGO2 and thus inhibits mRNA decay.
Hypoxia promotes the interaction of AGO2 with LUBAC. To investigate the underlying molecular mechanism, we analyzed proteins interacting with AGO2 in HeLa cells expressing Flag-AGO2 under hypoxia for 24 h by the mass spectrometry (MS). The interactions of AGO2 with two proteins HOIL-1L and HOIP, the components of LUBAC [16][17][18][19] , were identified, which were augmented by hypoxia (Fig. 2a, Supplementary Data 4). To validate whether AGO2 interacts with the components of LUBAC, lysates from 293T cells transfected with Myc-AGO2 and Flag-HOIP (Fig. 2b) or Flag-HOIL-1L (Fig. 2c) were used for coimmunoprecipitation (co-IP) with anti-Flag or anti-Myc antibody, respectively. The following immunoblotting results showed that AGO2 indeed interacted with HOIP and HOIL-1L (Fig. 2b, c). More convincingly, the association of endogenous AGO2 with HOIP, HOIL-1L and SHARPIN in HeLa cells were determined by co-IP with anti-AGO2 antibody (Fig. 2d). Since SHARPIN is also one of the components of LUBAC so we next confirmed the possibility of SHARPIN binding with AGO2. HeLa cell lysates were used for co-IP with anti-SHARPIN antibody, and the result showed that both HOIP and HOIL-1L were strongly but AGO2 weakly pulled down by SHARPIN (Fig. 2e). Taken together, all the above results illustrate that AGO2 interacts with LUBAC. More importantly, hypoxia greatly increased the interaction of endogenous of AGO2 with HOIL-1L and HOIP (Fig. 2f). To uncover why hypoxia augmented the interaction of LUBAC with AGO2, we investigated the expression level of LUBAC components in HeLa cells under hypoxia or in 293T cells treated with CoCl 2 , and found that hypoxia induced only HOIL-1L expression ( Supplementary Fig. 3a), which is consistent with reported previously 35,36 . Further, we confirmed that hypoxia induced the mRNA transcription of HOIL-1L but not of HOIP and SHARPIN by qRT-PCR ( Supplementary Fig. 3b). Next to examine whether HIF-1α is the transcript factor of HOIL-1L, the HOIL-1L promoter was subcloned to the luciferase reporter pGL3-basic to get an HOIL-1L luciferase reporter, which was transfected into 293T cell for 24 h, subsequently treated with hypoxia or CoCl 2 for the indicated times. The luciferase reporter results showed that hypoxia increased the HOIL-1L promoter activity (Supplementary Fig. 3c, d). Collectively, our results demonstrated that hypoxia induces the expression of HOIL-1L and thus increases the interaction of LUBAC with AGO2, which is consistent with the finding that deficiency of HOIL-1L lead to drastically destabilization of the LUBAC complex [16][17][18][19] .
To support that the increased association of HOIP with AGO2 is due to the more expression of HOIL-1L under hypoxia, Flag-HOIL-1L in a dose-increasing manner was transfected with the same amount of HA-AGO2 into 293T cells to mimic hypoxia condition. The co-IP/WB result showed that AGO2 recruited more endogenous HOIP along with increasing expression of HOIL-1L (Fig. 2g). On the contrary, knockdown of HOIL-1L enormously decreased the interaction of HOIP with AGO2 in HeLa cells (Fig. 2h). Since AGO2 is prolyl-4-hydroxylated at the residue proline 700 by P4H-α and P4H-β 37 , we investigated whether AGO2 prolyl-4-hydroxylation influences its interaction with LUBAC. 293T cells transfected with Flag-AGO2-WT or prolyl-4-hydroxylation mutant Flag-AGO2-P700A were treated with hypoxia for indicated times, cell lysates were used for co-IP with anti-Flag antibody. The results showed that hypoxia increased the interactions of AGO2-P700A with HOIP and HOIL-1L, as a similar pattern with AGO2-WT, indicating the prolyl-4-hydroxylation of AGO2 at P700 did not influence its interaction with HOIP and HOIL-1L under hypoxia stress (Fig. 2i). Moreover, to explore whether the induced HOIL-1L by hypoxia recruits more AGO2, HOIP or/and SHARPIN, HeLa cells treated with hypoxia for indicated times were lysed for co-IP with anti-HOIL-1L antibody, and then followed by Western blotting, which showed that hypoxia increased the association of HOIL-1L with HOIP, SHARPIN and AGO2 (Fig. 2j). We further examined the domains of HOIP or HOIL-1L that are involved in their protein-protein interaction with AGO2, Flag-tagged full length of HOIP or HOIL-1L and their domain deletion/truncated forms were co-transfected with HA-tagged AGO2 into 293T cells, then we used the anti-HA co-IP/WB to show that UBA domain of HOIP and UBL domain of HOIL-1L take the responsibility for the association with AGO2 (Fig. 2k, l), respectively. Taken together, all the above results illustrate that AGO2 interacts with LUBAC.
M1-linked linear ubiquitination of AGO2 is mediated by LUBAC in cells. LUBAC is the only enzyme which catalyzes the formation of linear poly-ubiquitin chain to its substrates, such as NEMO, RIPK1 17 , RIPK2 38 , ASC 39 42 . Recently, hundreds of linear polyubiquitinmodified substrates were identified through the mass spectrometry analysis combined with lysine-less internally tagged ubiquitin strategy 30 , indicating that M1-Ubi might participate in various cellular physiological processes. Such being in the case of AGO2 directly interacting with LUBAC which we have demonstrated above, we raised a question whether AGO2 is M1-linked linear poly-ubiquitinated. Firstly, we generated a mutant ubiquitin-7KR through mutation of all seven lysines of ubiquitin to arginines ( Supplementary Fig. 4a), so that the formation of poly-ubiquitin chain is a unique form of linear polyubiquitination linked by the first methionine (Met 1 ) rather than by lysine residues. Then we co-transfected AGO2 with or without ubiquitin-7KR into 293T cells, through IP with anti-Flag antibody and then immunoblotting by a specific linear poly-ubiquitin antibody LUB9 clone (or IE3 clone), and found that AGO2 was linear-ubiquitinated and this was increased by ectopically expressing ubiquitin-7KR ( Supplementary Fig. 4b). In addition, a Met 1 -linkage-specific ubiquitin binder (M1-SUB) 43 pull-down assay also showed a similar result ( Supplementary Fig. 4c).
Since AGO2 interacted with LUBAC, so we speculated that LUBAC is the E3 ligase of AGO2 for M1-Ubi. In order to confirm this, the components of LUBAC with ( Fig. 3a) or without ( Fig. 3b) myc-AGO2 were co-transfected into 293T cells, respectively. The M1-SUB pull-down assay find that LUBAC increased M1-Ubi of exogenous AGO2 (Fig. 3a). Meanwhile, M1-Ubi of endogenous AGO2 was also enhanced by LUBAC (Fig. 3b). To further support this, lysates from 293T cells ectopically expressing LUBAC complexes were co-incubated with purified GST-AGO2, subsequently pulled down by GST beads and immunoblotted by anti-LUB9, showing that LUBAC promoted M1-Ubi of AGO2 (Fig. 3c). Besides, M1-Ubi of endogenous AGO2 was augmented when HOIP and HOIL-1L were stably overexpressed in HeLa cells (Fig. 3d). More convincingly, we investigated LUBAC mediating M1-Ubi of AGO2 in vitro, and showed that LUBAC was capable of catalyzing M1-Ubi of AGO2 in vitro (Fig. 3e). These results demonstrate that AGO2 is a linear poly-ubiquitination substrate of LUBAC E3 ligases. Considering that SHARPIN is one of the essential components of LUBAC, we further investigated whether SHARPIN is necessary for M1-Ubi of AGO2. 293T cells stably knocking down SHARPIN were cotransfected AGO2 with HOIP and HOIL-1L, the M1-SUB pulldown assay showed that knockdown of SHARPIN slightly decreased M1-Ubi of AGO2 mediated by ectopic expression of HOIP and HOIL-1L ( Supplementary Fig. 4d). HOIP is the most crucial component of LUBAC, which catalyzes the formation of linear ubiquitin chain by its RING-IBR-RING (RBR) and extension LDD domains 44 . Stable knockdown of HOIP by shRNA in HeLa cells dramatically decreased the M1-Ubi of AGO2 (Fig. 3f). The sites of the cysteine at 885 in RING-IBR-RING (RBR) domain and the cysteine at 916 in the LDD domain are essential for HOIP enzymatic activity 44 , and the cysteine 460 is the key site for HOIL-1L E3 ligase activity. We co-transfected inactive mutant HOIP-C885A, HOIP-C916A or HOIL-1L-C460A with AGO2 into 293T cells for IP (Fig. 3g) and M1-SUB pulldown ( Supplementary Fig. 4e), the results revealed that M1-Ubi of AGO2 was dependent on the enzymatic activity sites C885 and C916 of HOIP but not C460 of HOIL-1L.
The RBR region of HOIP is sufficient to generate linear polyubiquitin chain in vitro whereas the full-length HOIP cannot, which is due to that the catalytic activity of HOIP is self-inhibited but can be released upon the LUBAC formation of HOIP with HOIL-1L, SHARPIN or both of them 45 . Gliotoxin (a fungal metabolite), which can block the enzymatic activity of LUBAC by selective binding to the RBR region of HOIP 46 , strongly decreased AGO2 M1-Ubi within half of 1 h and more apparently in 1 h ( Supplementary Fig. 4f). We also investigated the functional domain of HOIL-1L in mediating AGO2 M1-Ubi. Ubiquitin-like domain (UBL) of HOIL-1L is required for binding to HOIP to achieve the enzymatic activity, Npl4 type-zinc-finger domain (NZF) of HOIL-1L is required for binding to linear ubiquitin chain and RBR of HOIL-1L possesses E3 ubiquitin ligase. We cotransfected the functional domain truncated/deleted forms of HOIL-1L with HOIP and AGO2 into 293T cells for M1-SUB pull-down assay, the results showed that deletion of UBL domain other than NZF and RBR domains of HOIL-1L could not reshape HOIP catalytic activity to induce M1-Ubi of AGO2 (Supplementary Fig. 4g). Collectively, these above results demonstrate that LUBAC catalyzes M1-Ubi of AGO2.
OTULIN is a de-linear-ubiquitinase of AGO2. It has been known that OTULIN, CYLD, and A20 negatively regulate linear ubiquitin chain assembly via different mechanisms. OTULIN antagonizes LUBAC signaling highly specific hydrolyzing linear poly-ubiquitin chain 20,21,38 . CYLD is not only as a de-linearubiquitinase to remove linear poly-ubiquitin chain but also cleaving K63-linked poly-ubiquitin chain 23 . The C-terminal zincfinger 7 (ZF7) domain of A20 can directly bind to linear ubiquitin chain and impairs the interaction of LUBAC with substrates (such as NEMO) for M1-Ubi 22 . Considering that the highly specific de-linear-ubiquitin characteristic of OTULIN, we further investigated whether OTULIN is a de-linear-ubiquitinase of AGO2. HOIP, HOIL-1L and AGO2 were co-transfected with or Fig. 2 Hypoxia promotes the interaction of AGO2 with LUBAC. a Scatter distribution plot of proteins interacting with AGO2 under hypoxia and normoxia conditions were analyzed by mass spectrometry (MS). HeLa cells expressing Flag-AGO2 were treated with hypoxia or normoxia for 24 h, respectively. Cell lysates were used for co-IP with anti-Flag antibody, and followed by MS analysis. b, c Ectopically expressed Flag-HOIP or Flag-HOIL-1L interacted with AGO2. Myc-AGO2 was co-transfected with Flag-HOIP (b) or Flag-HOIL-1L (c) into 293T cells, the association of AGO2 with HOIP or HOIL-1L was determined by co-IP/WB. d HOIP, HOIL-1L and SHARPIN interacted with AGO2. Lysates from HeLa cells were used by co-IP with anti-AGO2 antibody, and followed by WB. e SHARPIN associated strongly with HOIP, HOIL-1L but weakly with AGO2. Lysates form HeLa cells were used by co-IP with anti-SHARPIN antibody, and followed by WB. f Hypoxia promoted the interaction of AGO2 with HOIL-1L and HOIP. HeLa cells treated with hypoxia were lysed for co-IP with anti-AGO2 antibody, and followed by WB with anti-HOIP and anti-HOIL-1L antibodies. g Increasing expression of HOIL-1L promoted AGO2 recruiting more LUBAC. Flag-HOIL-1L was transfected in a dose-increasing manner with HA-AGO2 into 293T cells. co-IP was performed with anti-HA antibody, and then endogenous HOIP and Flag-HOIL-1L bound to AGO2 were conducted by WB. h Knockdown of HOIL-1L decreased the interaction of AGO2 with HOIP. Stable HOIL-1L-knockdown HeLa cells were used for co-IP/WB with indicated antibodies. i AGO2 prolyl-4-hydroxylation did not influence its interaction with HOIL-1L and HOIP. 293T cells transfected with Flag-AGO2-WT or prolyl-4-hydroxylation mutant Flag-AGO2-P700A were treated with hypoxia for indicated times, and then lysed for co-IP/WB with indicated antibodies. j Hypoxia enhanced the association of HOIL-1L with HOIP and AGO2. HeLa cells treated with hypoxia for indicated times were lysed for co-IP/WB with indicated antibodies. k, l Analyses of HOIP and HOIL-1L binding domains with AGO2. Various truncated forms of Flag-HOIP (k) or Flag-HOIL-1L (l) were individually co-transfected with HA-AGO2 into 293T cells, and then lysates were used by co-IP with anti-HA antibody, followed by WB.
OTULIN cleaves linear ubiquitin chains via its PIM motif binding with the N-terminal PNGase/UBA (PUB) domain of HOIP. We generated HOIP mutants including N85/Y94A, Y82A, N102D, Y82A/N102D, which all interferes the association of HOIP with OTULIN. The M1-SUB pull-down assay showed that these HOIP mutants greatly promoted AGO2 M1-Ubi than that of wild-type HOIP ( Supplementary Fig. 4l). However, a relative lower expression level of OTULIN was enough to cleave the linear ubiquitin chain of AGO2, even such an augment of AGO2 M1-Ubi increased by HOIP-N85/Y94A, and this effect was not increased by the higher level of OTULIN ( Supplementary  Fig. 4m and OTULIN possessed a powerful role in de-linearubiquitination of AGO2. Moreover, since Y56 of OTULIN is conserved and critical for interacting with the PUB domain of HOIP and its phosphorylation negatively regulates this interaction 47 , we generated a mutant OTULIN-Y56F to abolish the phosphorylation for increasing the association of OTULIN with HOIP, which expectedly enhances de-linear-ubiquitination by OTULIN. However, de-linear-ubiquitination by wild-type OTULIN was not blocked by the addition of OTULIN-Y56F, which suggested a small amount of wild-type OTULIN was enough to cleave the linear ubiquitin chains on AGO2 ( Supplementary Fig. 4n). Thus, the above results demonstrate that OTULIN is a de-linear-ubiquitinase of AGO2.
K820 is one of M1-Ubi sites of AGO2 and hypoxia induces M1-Ubi of AGO2 catalyzed by LUBAC. To further identify M1-Ubi sites of AGO2, HeLa cells were treated with hypoxia for 12 or 24 h and then lysed in denaturing lysis buffer, which avoided pulldown of ubiquitin chains associated (not covalently modified) in IP with anti-AGO2 antibody. The IP-WB results showed that hypoxia strongly enhanced AGO2 M1-Ubi ( Supplementary  Fig. 5a) and the IP-MS analysis revealed that 19 lysine residues of AGO2 were ubiquitinated under hypoxic condition (Supplementary Fig. 5b). Moreover, lysates from 293T cells transfected with HA-AGO2, Flag-HOIP and Flag-HOIL-1L were also used for IP-MS analysis, revealing that AGO2 was multiply linearubiquitinated at 19 lysines ( Supplementary Fig. 5b). 12 lysine residues of AGO2 ubiquitination were overlapped among these residues identified as AGO2 ubiquitination sites under hypoxic condition and transiently transfection with Flag-HOIP and Flag-HOIL-1L by MS analysis (Supplementary Fig. 5b, c, Supplementary Data 5). We generated AGO2 functional domain constructs to probe its possible M1-Ubi sites, M1-SUB pull-down assay showed that M1-Ubi sites were mainly distributed in small RNA binding domains PAZ and MID as well as nuclease activity domain PIWI of AGO2 ( Supplementary Fig. 5d, e), implying that AGO2 was linear-ubiquitinated at multiple lysines. We have generated point or combined mutations for these lysines into arginines, and found that all those mutations could not completely abolish but the single-mutation K820R could reduce about a half of AGO2 M1-Ubi compared to that of AGO2-WT ( Fig. 3k; Supplementary Fig. 5f). Most importantly, considering that hypoxia induced the association of AGO2 with LUBAC (Fig. 2), we detected the dynamics of AGO2 M1-Ubi under hypoxia condition, showing that both hypoxia and CoCl 2 strongly enhanced AGO2 M1-Ubi (Fig. 3l). Taken together, these above results demonstrate that hypoxia induces M1-Ubi of AGO2 catalyzed by LUBAC.
AGO2 interacts with TNRC6/GW182 for miRISC assemble is promoted by phase separation 48 . Indeed, we found the cytoplasmic concentrating foci of miRISC formed by GFP-AGO2 and TNRC6A in HeLa cells ( Supplementary Fig. 6a). Moreover, the living HeLa cells showed a fusion of two adjacent GFP-AGO2 punctate foci after transient transfection of GFP-AGO2 (Supplementary Fig. 6b). Fluorescence recovery after photobleaching (FRAP) experiments displayed that plenty of the GFP-AGO2 molecules in punctate foci were quickly exchanged in the cytoplasm, and approximately 40% of GFP-AGO2 punctate foci fluorescence were recovered within 16 min ( Supplementary  Fig. 6c, d). Next, we wandered whether M1-Ubi of AGO2 induced by LUBAC regulates the processing of miRISC phase separation. HeLa cells stably overexpressing HOIP and HOIL-1L or knocking down HOIP were employed to obtain the high level and low level M1-Ubi of AGO2, respectively. The co-IP results showed that the AGO2 M1-Ubi levels did not influence the Fig. 3 M1-Ubi of AGO2 is catalyzed by LUBAC, which is induced by hypoxia. a LUBAC catalyzed M1-Ubi of AGO2. LUBAC related plasmids were transfected into 293T cells and subsequently performed by M1-SUB pull-down assay for detection of AGO2 M1-Ubi. b M1-Ubi of endogenous AGO2 was enhanced by LUBAC. 293T cells were co-transfected with components of LUBAC. Lysates were used for IP with anti-AGO2 antibody, followed by WB with a specific linear ubiquitin antibody 1E3 clone. c LUBAC catalyzed M1-Ubi of GST-AGO2 in vitro. Purified GST-AGO2 were incubated with lysates from 293T cells transfected with components of LUBAC, and followed by GST pull-down/WB assay with a specific linear ubiquitin antibody LUB9 clone. d LUBAC increased M1-Ubi of endogenous AGO2. Stable HeLa cells expressing HOIP and HOIL-1L were lysed for IP with anti-AGO2 antibody, and then determined M1-Ubi of AGO2 by WB with antibody 1E3 clone. e In vitro M1-Ubi assay for AGO2. Flag-AGO2 purified from 293T cells was co-incubated with purified ubiquitin and LUBAC proteins for in vitro M1-Ubi assay. f Knockdown of HOIP inhibited AGO2 M1-Ubi. Lysates from stable HeLa-shHOIP or -pLKO.1 cells were used for IP/WB. g AGO2 M1-Ubi was dependent on the catalytic activity of HOIP, but not that of HOIL-1L. 293T cells were cotransfected with indicated plasmids, and M1-Ubi of AGO2 was detected by IP/WB. h M1-Ubi of AGO2 was removed by OTULIN. 293T cells were cotransfected with indicated plasmids, and M1-Ubi of AGO2 was analyzed by M1-SUB pull-down/WB. i Knockdown of OTULIN increased AGO2 M1-Ubi. Stable HeLa-shOTULIN cells were transfected with indicated plasmids, and AGO2 M1-Ubi was analyzed by M1-SUB pull-down/WB. j OTULIN mutants (W96A, C129A) did not inhibit AGO2 M1-Ubi. OTULIN or indicated OTULIN mutants were transfected into 293T cells, respectively. M1-Ubi of AGO2 was analyzed by IP/WB. k K820 is a major site for M1-Ubi of AGO2. The M1-Ubi of AGO2-WT and indicated AGO2 mutants were detected by M1-SUB pulldown and WB. l Hypoxia-induced M1-Ubi of AGO2. 293T and HeLa cells were treated with hypoxia or CoCl 2 (300 μM) for indicated times, and AGO2 M1-Ubi was determined by IP/WB. interaction of AGO2 with TNRC6A ( Supplementary Fig. 6e, f). Consistently, the number of GFP-AGO2/TNRC6A colocalization foci per cell was not influenced when HOIP and HOIL-1L were stably overexpressed or OTULIN was knocked down by shRNA in HeLa cells ( Supplementary Fig. 6g, h). Conversely, decreased AGO2 M1-Ubi through knocking down of HOIP by shRNA in HeLa cells did not affect the number of GFP-AGO2/HA-TNRC6C co-localization foci per cell ( Supplementary  Fig. 6i). These results illustrate that M1-Ubi of AGO2 is not involved in the processing of miRISC phase separation. Taken together, all the above results demonstrate that HOIP and HOIL-1L are co-localized with miRISC.  The scale of the magnified region (white frame) is 2 μm x 2 μm (e, g). Co-localization between GFP-AGO2/TNRC6A with HOIP or HOIL-1L were analyzed by Pearson's correlation coefficient. Data were mean ± s.d., and P-values were determined by unpaired two-sided t-test (f, h). transfected with indicated plasmids were harvested for the psiCHECK2-4xlet-7a-BS of dual-luciferase assay. Dual-luciferase reporter assay data were mean ± s.e.m., n = 3 (a, c, d) or n = 4 (b) biologically independent experiments, and P-values were determined by unpaired two-sided t-test. e, f Knockdown of HOIP increased let-7a-targeted GFP protein and mRNA levels. HeLa cells were transfected with indicated plasmids and siRNAs, the GFP protein expression was detected by WB (e). 293T cell stably expressing GFP-4x-let-7a-BS was transfected with HOIP siRNAs, and then total RNAs were extracted for detection of GFP mRNA levels by qRT-PCR. Data were mean ± s.d., n = 4 biologically independent samples, and P-values were determined by unpaired two-sided t-test (f). g, h Knockdown of OTULIN decreased let-7a-targeted GFP and c-MYC decay. HeLa cells were transfected with indicated plasmids and siRNAs, the proteins GFP (g) and c-MYC (h) expression were measured by WB analysis.
To further support this, we co-transfected GFP-4xlet-7a-BS, HOIP siRNA with or without HA-AGO2 into HeLa cells. As expectedly, the GFP expression level was suppressed AGO2, and this effect was much stronger when HOIP was knocked down by siRNA (Fig. 5e). Due to AGO2/miRNA-mediating target gene silencing on the post-transcript level, so we investigated the mRNA level measured by QRT-PCR in 293T cells stably expressing GFP-4xlet-7a-BS which was co-transfected with AGO2 and HOIP siRNAs, to show that knockdown of HOIP increased the let-7a miRISC activity for GFP mRNA decay (Fig. 5f). In contrast, knockdown of OTULIN by siRNAs suppressed the let-7 miRISC activities for silencing of targeted GFP (Fig. 5g) and endogenous c-MYC (Fig. 5h). Taken together, these data demonstrate that M1-Ubi of AGO2 undermines miRNA-mediated gene silencing.
M1-Ubi of AGO2 does not influence miRNA processing. Next, we tried to explore the mechanism underlying M1-Ubi of AGO2 affecting the miRNA pathway. Knockdown of either HOIP or OTULIN to reshape the status of M1-Ubi of AGO2 did not affect the expression levels of core proteins in the miRNA pathway, such as DICER, AGO2 and TARBP2 (Supplementary Fig. 7a, b). Further, Myc-AGO2 and Flag-HA-DICER were co-transfected into HeLa cells in which HOIP was knocked down by shRNA, the co-IP/WB results showed that knockdown of HOIP does not influenced the interaction of AGO2 with DICER ( Supplementary Fig. 7c). Moreover, AGO2 was ectopically expressed with HOIL-1L, HOIP or HOIPmutants in 293T cells to show that the association of AGO2 with endogenous DICER was not affected by overexpression of either HOIP/HOIL-1L or HOIP-C885A/HOIL-1L and HOIP-C916A/HOIL-1L mutants (Supplementary Fig. 7d). These results demonstrate that M1-Ubi of AGO2 has no role in regulating the expression level of core proteins in the miRNA pathway and the binding of AGO2 with DICER.
To estimate whether M1-Ubi of AGO2 participates in the pre-miRNA processing, we employed an in vitro pre-miRNA processing assay. Lysates from 293T cells transfected with indicated plasmids were immunoprecipitated by HA-AGO2, then the beads were co-incubated with purified biotin-tagged pre-let-7a-3 for the following in vitro processing assay. The results revealed that AGO2 effectively promoted the mature let-7a biogenesis, but overexpression of HOIP-WT, -C885A or -C916A and HOIL-1L had no effect on mature let-7a biogenesis processed by AGO2 (Supplementary Fig. 7e), which suggested that M1-Ubi of AGO2 did not influence the pre-let-7a-3 processing. AGO2 possesses the nuclease activity for cleavage of targeted mRNA and passenger strand of small RNA duplexes 3 , which is essential for biogenesis of several miRNAs, such as miR-451 49,50 . Pri-miR-451a and HOIP siRNAs or OTULIN siRNAs were co-transfected with or without AGO2 into 293T cells, and then Northern blotting analysis was performed for assessment of the nuclease activity of AGO2 by the expression level of mature miR-451. As expectedly, AGO2 could remarkably increase the biogenesis of miR-451a; however, it was not influenced by knockdown of either HOIP or OTULIN ( Supplementary Fig. 7f, g).

M1
-Ubi of AGO2 impedes its recruitment of miRNA-targeted mRNAs. We speculated the mechanism for M1-Ubi of AGO2 disrupting the miRISC activity was probably due to the retard of target mRNA recruitment to AGO2. The miRNA profiles by small RNA-Seq in stable HeLa-Flag-AGO2 cells expressing HOIP and HOIL-1L and HeLa cells knocking down HOIP were analyzed to find that a portion of miRNAs were changed (Fig. 6a, b,  Supplementary Data 2). Then we performed RIP-Seq with anti-Flag and anti-AGO2 antibody in these two group stable cell lines, respectively. A total of 7014 and 5788 mRNA transcripts associated with AGO2 were identified as overlapping between Control (9595, 73.1%), HOIP + HOIL-1L (9576, 73.2%), pLKO.1 (9060, 63.9%) and shHOIP (12118, 47.8%) cells, respectively (Supplementary Fig. 8a, b). RIP-Seq results showed that overexpression of LUBAC reduced mRNA transcripts bound to AGO2 (P = 1.053E−8, Mann-Whitney U test) (Fig. 6c, Supplementary Data 6), whereas knockdown of HOIP significantly increased the interaction of mRNA transcripts with AGO2 (P = 1.912E-102, Mann-Whitney U test) (Fig. 6d, Supplementary Data 7). Distribution plots analysis showed that the suppressed AGO2 bound mRNA transcripts (1131, 2-fold up-regulated) were 1.53 fold more than the increased (739, 2-fold down-regulated) by ectopic expression of HOIP and HOIL-1L ( Supplementary Fig. 8c). On the contrary, the increased AGO2 bound mRNA transcripts (2221, 2-fold up-regulated) were 2.92 fold more than the reduced (760, 2-fold down-regulated) by knocked down of HOIP (Supplementary Fig. 8d). Combining analysis of above RIP-Seq data displayed 1020 mRNAs associated with AGO2 in all these cell lines (Fig. 6e). Notably, most of these mRNA transcripts bound to AGO2 were significantly reduced in HOIP and HOIL-1L overexpressing cells and simultaneously increased in HOIP knocking down cells (Fig. 6f). To further confirm this, RIP-qRT-PCR was measured to reveal that overexpression of LUBAC significantly decreased and knockdown of HOIP enormously augmented the association of several mRNA transcripts including NOMO1, TBC1D20, GFPTA, RAPH1 and PUM1 with AGO2, respectively (Fig. 6g, h).
To make sure that the impaired recruitment of targeted mRNA to M1-Ubi of AGO2 was mediated by miRNA. We analyzed the miRNA expression profile by miRNA-Seq in this two group cell lines (Supplementary Data 2). Comparative analyze to RIP-Seq data, the cumulative fraction results showed that overexpression of HOIP and HOIL-1L significantly inhibited the association of 2792 top-10 non-difference miRNA-targeted mRNAs with AGO2 (Fig. 6i); on the contrary, knockdown of HOIP greatly promoted the recruitment of 982 top-10 non-difference miRNA-targeted mRNAs to AGO2 (Fig. 6j). Consistently, cataloging targets by miRNA-seed complementary sequence showed that overexpression of HOIP and HOIL-1L reduced the interaction of complementarily paired target mRNAs with 6mer (n = 2310), 7mer (n = 383) and 8mer (n = 99) sequences of top-10 nondifference miRNAs to AGO2, respectively ( Supplementary  Fig. 8e). Conversely, knockdown of HOIP increased the association of complementarily paired target mRNAs with 6mer (n = 816), 7mer (n = 127) and 8mer (n = 39) sequences of top-10 non-difference miRNAs with AGO2, respectively ( Supplementary  Fig. 8f). These results demonstrate that M1-Ubi of AGO2 impairs its recruitment of miRNA-targeted mRNAs.
In light of the above results, we speculated that the mRNA transcript half-life for decay may be influenced by M1-Ubi of AGO2. Transcripts bound to AGO2 were classified on the basis of published transcriptome-wide half-life measurements 51 , in which the half-life during 0-4 h, 4-12 h and >12 h were categorized as short, middle and long, respectively. Indeed, the abundance of mRNA transcripts with a longer-half-life were significantly much more than those with a shorter-half-life when HOIP and HOIL-1L were expressed (Fig. 7c), suggesting that M1-Ubi of AGO2 critically regulated the half-lives of mRNA transcripts. To further investigate this, HeLa-shHOIP cells were transfected with GFP-4xlet-7a-BS, and then treated with Actinomycin D for indicated times, the expression level of GFP mRNA was measured by qRT-PCR to show that knockdown of HOIP greatly shortened the half-lives of GFP mRNA (Fig. 7d). Moreover, GO analysis showed that several distinct gene clusters such as translation, DNA repair, cell cycle, mRNA processing and cell division were gathered when HOIP and HOIL-1L were ectopically expressed (Fig. 7e), which coincided with the GO enrichment terms in HeLa cells under hypoxia ( Supplementary Fig. 2m). Taken together, our results suggest that M1-Ubi of AGO2 globally inhibits mRNA decay. In addition, we also analyzed the lncRNA expression profiles and found that the increased lncRNA transcripts (186, FPKM > = 30) were 9.3 fold more than the decreased (20, FPKM > = 30) by ectopic expression of HOIP and HOIL-1L (Fig. 7f, Supplementary  Data 3), suggesting that M1-Ubi of AGO2 regulates turnover of not only mRNAs but also lncRNAs.
Hypoxia promotes accumulation of miRNA-targeted mRNAs in clinical tumor specimens. To investigate whether hypoxiainduced M1-Ubi of AGO2 promoting mRNA accumulation is also an ubiquitous phenomenon in clinical tumor specimens, TCGA data of miRNA-Seq and RNA-Seq in lung cancer samples were employed. Firstly, we compared the expression levels of miRNAs in lung cancer tissues (n = 999), where are regarded as a relative higher hypoxia degree, with those of normal tissues (n = 91), to find that the top 10 non-difference miRNAs including mir-22, mir-10a, mir-103a-1, mir-103a-2, mir-92a-1, mir-92a-2, mir-23b, mir-142, mir-146b and mir-24-2 were almost not changed between normal tissues and lung cancer tissues (Fig. 8a, b). Further, RNA-Seq data in lung cancer samples (n = 1028) were analyzed to estimate the hypoxia degree by using the method of GSVA (Gene set enrichment analysis) 52 with a hypoxia metagene signature 53 . The results showed that hypoxia scores were very positively correlated with the expression levels of the top 10 nondifference miRNA-targeted mRNAs (Pearson r = 0.7718, P value (two tailed) < 0.0001(****)) (Fig. 8c, upper panel) whereas slightly negative correlated with the expression levels of the non-miRNA targets (Pearson r = −0.07952, P value (two tailed) = 0.0108(*)) ( Fig. 8c, lower panel). Moreover, we divided these lung cancer samples as two groups, low-hypoxia (hypoxia score < 0, n = 457) and high-hypoxia (hypoxia score > 0, n = 571), according to the hypoxia score. The results showed that the expression levels of the top 10 non-difference miRNA targets in high-hypoxia lung cancer tissues were much higher than those in low-hypoxia lung cancer  HOIP (b). c, d M1-Ubi of AGO2 inhibited recruitments of mRNAs. Cumulative fraction analyses for mRNA transcripts recruiting to AGO2 were conducted by using RIP-Seq data in stable HeLa-Flag-AGO2 cells expressing HOIP and HOIL-1L (c, n = 7014 mRNA transcripts) and HeLa cells knocking down HOIP (d, n = 5788 mRNA transcripts). e, f Venn diagram analysis (e) and scatter plots analysis of mRNA transcripts recruited to AGO2 with more than 1.5-fold changes (f). g, h Several mRNAs bound to AGO2 were examined by RIP/qRT-PCR in above stable HeLa cells. i, j Cumulative fraction analyses for top 10 non-difference miRNA-targeted mRNA transcripts recruiting to AGO2 (i, n = 2792 mRNA transcripts; j, n = 982 mRNA transcripts) were conducted by using RIP-Seq data in stable HeLa cells. k LUBAC inhibit the association of let-7-targeted HMGA2 and c-MYC mRNAs with AGO2. HMGA2 and c-MYC mRNAs bound to AGO2 were measured by RIP/qRT-PCR. The RIP efficiency were determined by WB and Northern blotting analysis. l-n Knockdown of HOIP augmented the associations of let-7a-targeted mRNAs to AGO2. HeLa-shHOIP cells transfected with indicated plasmids were used for target mRNA pull-down assay (l) and GST-MS2 pull-down assay (m, n). l RNA bound fraction (beads) and unbound fraction (supernatant) were detected by WB. m A schematic illustration of the GST-MS2 pull-down assay. n AGO2 associated RNAs through let-7a were assessed by GST-MS2 pull-down. o K820R of AGO2 augmented its recruiting with let-7-targeted mRNAs. The abundance of let-7a and let-7a targeted HMGA2/c-MYC mRNAs associated with AGO2 or mutants were determined by RIP/northern blotting and RIP/ qRT-PCR, respectively. qRT-PCR data were mean ± s.d., n > = 3 biologically independent samples, and P-values were determined by unpaired two-sided ttest (g, h, k, o). In box plots, the lines represent the median, first and third quartiles, the whiskers denote the minima and maxima; P-values were calculated using a two-sided Mann-Whitney U test for cumulative fraction analysis (c, d, i, j). tissues (Fig. 8d, left panel), but the expression levels of the non-miRNA targets were not much different between high-hypoxia and low-hypoxia lung cancer tissues (right Fig. 8d). Collectively, these results demonstrate that hypoxia scores in clinical tumor specimens are positively correlated with the expression levels of miRNA-targeted mRNAs.

Discussion
Our results find that hypoxia can lead to increase in the expression levels of thousands and thousands of miRNA-targeted mRNAs in tumor cells; however, none of the currently known molecular mechanisms can well explain it. One classical model is that the transcription factor HIF-1α upregulation induced by hypoxia, but which gives rise to the transcriptions of only more than 100 downstream genes in cells 11 . Accordingly, we have raised a key scientific question: How does hypoxia in tumor cells regulate overall mRNA degradation and homeostasis, and how does this regulation play a key role in tumor phenotype? In this study, we answer those questions and reveal a mechanism that hypoxia in the tumor microenvironment is an incentive for  mRNA turnover through inducing M1-Ubi modification of the key protein AGO2 in controlling the efficiency of the miRNA pathway (Fig. 8e). Briefly, hypoxia induces the expression of HIF-1α, which up-regulates the transcription of HOIL-1L, the core component of LUBAC. The increased HOIL-1L with its partner HOIP is recruited to and catalyze AGO2 for M1-Ubi, which promotes AGO2 to form a functional hindered rather than normal miRISC. Subsequently, miRNA-targeted mRNAs cannot be loaded to the M1-Ubi modified AGO2/miRISC, thus the overall miRNA-targeted mRNAs are not degraded and stabilized in hypoxic tumor cells. By this mechanism, short-term hypoxia stress may protect overall mRNAs and enhances stress tolerance, whereas long-term hypoxia stress as like in tumor cells results in serious changing of the entire gene expression profile, which is a driving force in the dynamic process of cell malignant evolution. In addition, we find that M1-Ubi of AGO2 can be de-linearubiquitinated by OUTLIN.
Although there are many literatures in M1-Ubi, LUBAC and its related substrates are almost focused on inflammation and immune signaling. Increasing evidence suggest that M1-Ubi is involved in non-inflammatory and non-immunological signaling and other functions. LUBAC is recruited to TDP-43 and SOD1 of Htt-polyQ aggregates thus mediating their M1-Ubi, which plays a great role in regulating intracellular stability and proteotoxicity of these proteins implicated in neurodegenerative disease 41 . M1-Ubi also plays a role in controlling chromosome alignment during mitosis, in which LUBAC catalyzes M1-Ubi of the kinetochore motor CENP-E and then regulates its localization at attached kinetochores for the following chromosome alignment 42 . Furthermore, M1-Ubi of STAT1 can inhibit its binding to the type-I interferon receptor IFNAR2, thereby restricting STAT1 activation and controlling antiviral interferon signaling 40 . Here we reveal the function of M1-Ubi in controlling miRNA-targeted mRNA recruiting to AGO2 and modulating mRNA accumulation under hypoxia. Previous researches have revealed that M1-Ubi at different substrates implement diverse functions by mainly changing the protein-protein interactions 17,38,39 . Interestingly, we demonstrate that M1-Ubi acts as a crucial mode in the miRNA pathway through influencing the RNA-protein interaction (which is mRNA recruitment to AGO2 protein) under hypoxia.
The active miRISC formation includes several steps: first, miRNA is loaded to AGO2; second, miRNA-targeted mRNA is recruited to AGO2/miRNA by 3′-untranslated region complementarily pairing with miRNA; third, TNRC6/GW182 is then bound to the complex of AGO2/miRNA/mRNA in a manner of phase separation and the active miRISC is finally established for mRNA silencing. Several factors participate in this process: (1) some RNA binding proteins (RBP) associate directly with miRISC or bind to the same targeted mRNA adjacent to the miRISC location. For examples, HuR, an AU-rich element-binding protein, interacts with cationic amino acid transporter 1 (CAT-1) mRNA near to the miR-122-RISC location and inhibits its activity under amino acid starvation 54 . FUS promotes the activity of miRNA-mediated gene silencing through direct association with the miRISC components including AGO2, miRNAs and targeted mRNAs, for optimal miRNA-mediated gene silencing 55 .
(2) Co-factor proteins directly interact AGO2, such as nucleoporin Nup358 interacts with AGO2 and promotes the association of targeted mRNA with AGO2/miRNA 56 . Eukaryotic translation initiation factor (eIF1A) interacts with AGO2 and augments AGO2-mediated DICER-independent miRNA biogenesis and miRNA-mediated gene silencing 57 . (3) Post-translational modifications of AGO2 also take responsibility for this process. S-Nitrosylation of AGO2 at Cys691 by microbiome-derived NO increases its interaction with TNRC6/GW182 for miRNAmediating gene expression 58 . Phosporylaiton of AGO2 at Tyr529 impairs the miRNA loading 59 , and this tyrosine residue within the 5′-phosphate-binding pocket of the MID domain has been found directly associated with the first nucleotide phosphate group of miRNA on the molecular crystal level 60 . Akt-mediated phosphorylation of AGO2 at S387 inhibits its association with GW182 and then suppresses miRNA-mediated gene silencing 61 . Phosphorylation of AGO2 at S824-S834, located at a structurally unresolved loop of PIWI domain 60 , impairs its interaction with target mRNA and subsequently suppresses miRNA-mediated gene silencing 62,63 . It is important to find a unrevealed modification of the core molecule AGO2 elucidating the mechanism under specific cellular conditions, such as hypoxia. We identify AGO2 can occur M1-Ubi modification, which is majorly and functionally at K820 that is very close to the phospho-site region S824-834. The M1-Ubi modified AGO2/miRISC is functionally hindered to recruit miRNA-targeted mRNAs, thus blocking the overall mRNA decay.  Analysis of all (a) and top 10 (b) non-difference miRNAs between normal tissues (n = 91) and lung cancer tissues (n = 999) from TCGA lung cancer miRNA-seq data. c The expression levels of the top 10 non-difference miRNA-targeted mRNAs were positively correlated (Pearson r = 0.7718, P value (two tailed) < 0.0001(****)) and the non-miRNA targets were slightly negative correlated (Pearson r = −0.07952, P value (two tailed) = 0.0108(*)) with hypoxia scores from TCGA lung cancer (n = 1028) RNA-Seq data, respectively. d The expression levels of the top 10 non-difference miRNA targets and the non-miRNA targets in high-hypoxia lung cancer tissues were much higher and slightly lower than those in low-hypoxia lung cancer tissues, respectively. Clinical lung cancer samples were divided into two groups, low-hypoxia (hypoxia score < 0, n = 457) and high-hypoxia (hypoxia score > 0, n = 571), according to the hypoxia score. Data were mean ± s.d., and Pvalues were determined by unpaired two-sided t-test. e Mechanistic model summarizing how hypoxia in the tumor microenvironment was an incentive for mRNA accumulation by inducing AGO2 M1-Ubi. Briefly, Hypoxia induces HOIL-1L transcription and enhances LUBAC interacting with AGO2, thereby catalyzing M1-Ubi of AGO2, which can be conversely removed by OUTLIN. M1-Ubi of AGO2 suppresses miRNA-targeted mRNAs loading to AGO2, thus to inhibit the miRISC activity, consequently augment global-mRNA accumulation.
Determination of M1-Ubi of AGO2 under denaturing condition. HeLa cells treated with hypoxia as indicated times were lysed with denaturing lysis buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 2% SDS, 5 mM DTT and Protease inhibitor cocktail) for 10 min, and then lysates were boiled at 95°C for 10 min and sonicated. Subsequently, cell lysates were centrifuged at 16,000 g for 30 min at 4°C, then supernatant lysates were diluted 1:20 with lysis buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 5 mM DTT, 1% NP-40 and protease inhibitor cocktail) for the following immunoprecipitation (IP) with anti-AGO2 antibody and incubated with Protein A/G beads overnight at 4°C. Beads were washed five times with lysis buffer without protease inhibitor cocktail, and then dissolved in SDS-PAGE protein loading buffer containing DTT and denatured at 95°C for 10 min, a part of the samples were measured by Western blot for AGO2 M1-Ubi analysis. Meanwhile, the remained samples were subjected to SDS-polyacrylamide gel for coomassie brilliant blue staining, the gels containing linear ubiquitinated AGO2 were cut and digested with trypsin for mass spectrometry analysis.
Co-immunoprecipitation (Co-IP). 293T or HeLa cells transfected with indicated plasmids were lysed with RIPA lysis buffer (50 mM Tris-HCl pH7.4, 150 mM NaCl, 1% NP-40, protease inhibitor cocktail). Cell lysates were incubated with indicated antibodies and protein A/G-agarose beads at 4°C overnight, and then washed for five times with RIPA lysis buffer without protease inhibitor cocktail and followed by Western blotting analysis.
M1-SUB pull-down assay. 293T cells were transfected with indicated plasmids or treatment of hypoxia, and then cells were lysed with RIPA lysis buffer (50 mM Tris-HCl pH7.4, 150 mM NaCl, 1% NP-40, protease inhibitor cocktail) on ice for 1 h. Lysates were incubated with purified GST-UBAN protein and Glutathione sepharose 4B beads at 4°C overnight. Linear ubiquitinated AGO2 bound to GST-UBAN-beads were washed for three times by using RIPA lysis buffer, and followed by Western blotting analysis.
RNA immunoprecipitation (RIP) and qRT-PCR analysis. The methods were modified from our previous study 64 . Briefly, stable cell lines were treated with or without hypoxia, and then were lysed in RIP-lysis buffer (50 mM Tris-HCl pH7.4, 150 mM NaCl, 10 mM EDTA, 5 mM MgCl 2 , 1% NP-40, 1 mM DTT, 100 units/ml RNase inhibitor (Fermentas), 400 μM VRC (New England BioLabs) and Protease inhibitor cocktail) on ice for 1 h. 1/10 of lysates were extracted with 1 ml of TRIZOL reagent (Invitrogen) for extraction of total RNAs as input, 1/50 of lysates were saved for Western blotting to detect the protein expression, and left lysates were incubated with indicated antibodies and protein A/G-agarose beads at 4°C overnight. Beads bound with RNAs were washed with the same RIP-lysis buffer for five times, then 1/10 of beads was subjected to Western blotting analysis to identify the efficiency of IP, the remained beads were extracted with 1 ml of TRIZOL reagent for extraction of RIP bound RNAs.
For RIP-qRT-PCR, RNAs extracted from RIP bound or total input RNAs were firstly treated with DNase I (Thermo) to degrade genomic DNA. Reverse transcription was performed by using PrimeScript TM RT-PCR Kit (TAKARA), oligo-dT primers and random primers were used to reverse transcript. SYBR ® Green PCR Master Mix (Applied Biosystems) was used for qPCR on StepOne Plus Real-Time PCR System (Applied Biosystems) to analyze the RNAs bound to AGO2; RIP fractions of mRNA were normalized and calculated as % input.
Let-7a luciferase reporter for miRISC activity assay. Four-repeated let-7a binding site (BS) sequences of 4xlet-7a-BS were inserted into the 3′-UTR of Renilla luciferase of the vector psiCHECK2 to constitute psiCHECK2-4xlet-7a-BS. Accordingly, Renilla luciferase is expectedly attenuated when let-7a binds to binding sites. AGO2, psiCHECK2-4xlet7a-BS and indicated plasmids or siRNAs were co-transfected into 293T cells, 48 h after transfection the cells were subjected to the dual-luciferase reporter assay according to the manufacturer's instruction. The miRSIC activity with Renilla luciferase was normalized by Firefly luciferase.
GFP expression for miRSIC activity assay. Four-repeated let-7a binding site sequences of 4xlet-7a-BS was inserted into the 3′UTR of GFP in the vector of pEGFP-C1 to get pEGFP-C1-4xlet-7a-BS, and then the reporter pEGFP-4xlet-7a-BS was subcloned into the lentiviral vector pCD510 to form pCD510-GFP-4xlet-7a-BS for construction of stable cell lines. HeLa or 293T cells were transfected with pEGFP-C1-4xlet-7a-BS and indicated plasmids or siRNAs, 48 h after transfection cells were collected for Western blotting and qRT-PCR analyses of GFP in protein level and mRNA level, respectively.
HOIL-1L promoter luciferase activity assay. The HOIL-1L promoter region (2.0 kb upstream of transcription start site) was cloned into the vector pGL3-Basicluciferase reporter (Promega) to get pGL3-HOIL-1L-P. 293T cells were seeded on 24-well plates and co-transfected with the Renilla luciferase reporter pSV-40 and pGL3-HOIL-1L-P or pGL3-Basic. 48 h later the luciferase activity was measured by dual-luciferase reporter assay according to the instruction, the HOIL-1L promoter luciferase was normalized by Renilla luciferase activity.
Mass spectrometry (MS) analysis. For identification of the proteome profile associated by AGO2 under hypoxia, HeLa cells stably expressing Flag-AGO2 were treated with normoxia (21% O 2 ) and hypoxia (1% O 2 ) for 24 h, respectively. For identification of potential M1-Ubi sites of AGO2, 293T cells were co-transfected with HA-AGO2 and Flag-HOIP/HOIL-1L for 48 h. Cells were lysed with RIPA lysis buffer (50 mM Tris-HCl pH7.4, 150 mM NaCl, 10 mM EDTA, 1% NP-40, protease inhibitor cocktail tablet), then lysates were incubated with anti-Flag M2 monoclonal antibody or anti-HA and protein A/G-agarose beads at 4°C overnight. Beads were washed for five times with the same RIPA lysis buffer, and then dissolved in 10% SDS lysis buffer, denatured at 95°C for 10 min. For the proteome profile, the samples were directly used for trypsin digestion step; for the M1-Ubi sites, the samples were subjected to SDS-polyacrylamide gel for coomassie brilliant blue staining, the gel containing linear ubiquitinated AGO2 was cut for trypsin digestion step.
For mass spectrometry analysis, Nano-LC-MS/MS was used 65 . Protein samples were gradually reduced and alkylated by dithiothreitol (DTT) in a final concentration of 10 mM at room temperature for 1 h and iodoacetamide in a final concentration of 55 mM at room temperature for 1 h in the dark conditions, respectively. Next, the samples were digested by sequencing-grade modified trypsin (Promega) at 37°C overnight, and then collected by centrifugation for 20 min at 14,000 g at room temperature. The tryptic peptides were treated with 1% trifluoroacetic acid, purified by using C18 Ziptips and eluted with 0.1% trifluoroacetic acid in 60% acetonitrile. The eluted peptides were lyophilized using a SpeedVac (ThermoSavant), and resuspended in 10 μl of 1% formic acid/5% acetonitrile. All mass spectrometric experiments were performed on an Orbitrap Fusion LUMOS mass spectrometer (Thermo Fisher Scientific) connected to an Easy-nLC 1200 via an Easy Spray (Thermo Fisher Scientific). The reverse-phase microcapillary column (0.1 × 150 mm) packed with Reversed Phase C18 resins (2 μm, PepMap RSLC) was used for mixed peptides loading at a flow of 1 μl/min. Then a 60-min linear gradient solution from 95% buffer A (0.1% formic acid, 2% acetonitrile and 98% water) to 30% buffer B (0.1% formic acid and 80% acetonitrile) at a flow rate of 0.3 μl/min was conducted for peptides reverse-phase separation. The radio frequency (RF) lens was 60%, mass spectrometer was operated in positive ion mode and employed in the data-dependent mode within the specialized cycle time (3 s), to automatically switch between mass spectrometry and tandem mass spectrometry (MS/MS). One full mass spectrometry scan from 350-1, 500 m/z was acquired at high resolution (R = 120,000 (defined at m/z = 400)). MS/MS scans were acquired at a resolution of 30,000. Masses selected for MS/MS were isolated (quadrupole) at a width of 4 Da and were fragmented using a higher energy collisional dissociation of 30% ± 5. All MS/MS ion spectra were analyzed by using PEAKS 10.0 (Bioinformatics Solutions) for processing, de novo sequencing and database searching. Resulting sequences were searched against the UniProtHuman Proteome database. Carbamidomethyl (57.02 Da) of cysteine was specified as a fixed modification, oxidation (15.99 Da) of methionine, acetylation (42.01 Da) of the N terminus and ubiquitination (114.04 Da) of lysine were specified as variable modifications. FDR estimation was enabled. Peptides and proteins were separately filtered for −log10(P value) ≥ 20 and for −log10(P value) ≥ 15. For all of the experiments, this gave an FDR < 1% at the peptide-spectrum match level. Proteins sharing significant peptide evidence were grouped into clusters. The relative ratios (hypoxia/normoxia) of proteins associated with AGO2 under hypoxia and normoxia conditions were calculated by normalizing to AGO2 itself. The mass spectrometry (MS)-based proteomics was performed at the Proteomics of Core Facility of Basic Medical Sciences, Shanghai Jiao Tong University School of Medicine (SJTU-SM).
Immunofluorescence staining. For immunofluorescence (IF) staining 66 , HeLa cells, stable HeLa cells expressing of HOIP/HOIL-1L and HeLa cells knocking down of HOIP or OTULIN were cultured on glass coverslips in 24-well plates, and then transfected with indicated plasmids. 24 h after transfection the cells on coverslips were washed twice times with PBS and immobilized by 4% PFA for 30 min at room temperature. After washing three times with PBS (5 min each time), cells were blocked by blocking solution (5% BSA in TBS with 0.5% Triton X-100) for 1 h for the following incubation with antibodies against TNCR6A/GW182, HOIP, HOIL-1L and HA in blocking solution for 1 h at room temperature. Moreover, cells were washed three times with PBS (5 min each time) and then incubated with appropriate secondary antibodies conjugated with AlexaFluorescence 568 or with AlexaFluorescence 647 in blocking solution for 1 h at room temperature away from light. Furthermore, after washed three times with PBS (5 min each time), the cells were stained with DAPI for 30 min in the dark. Finally, IF images were captured using a laser scanning confocal microscopy.
High-throughput sequencing for RIP-Seq, RNA-Seq and miRNA-Seq. For RIP-Seq, indicated stable HeLa cell lines were conducted by RNA immunoprecipitation (RIP) for the following sequencing were performed as described 67 . RNAs bound to AGO2 and total RNAs (as an input) were extracted by using TRIZOL reagent as following manufacturer's instruction (Invitrogen), then the rRNAs were removed from the immunoprecipitated RNA and input RNA samples by using RNAs with NEBNext rRNA Depletion Kit (New England Biolabs, Inc., Massachusetts, USA). The rRNA-depleted RNAs were constructed RNA sequencing libraries by using NEBNext ® Ultra™ II Directional RNA Library Prep Kit (New England Biolabs, Inc., Massachusetts, USA) according to the manufacturer's instructions.
For RNA-Seq, total RNA (1 μg) extracted from indicated stable HeLa cell lines by TRIZOL reagent was used for removing the rRNAs using Ribo-Zero rRNA Removal Kits (Illumina, San Diego, CA, USA) as the manufacturer's instructions for the following library construction. The rRNA-depleted RNAs were constructed RNA sequencing libraries by using TruSeq Stranded Total RNA Library Prep Kit (Illumina, San Diego, CA, USA) according to the manufacturer's instructions.
For miRNA-Seq, total RNAs were extracted from indicated stable HeLa cell lines by using TRIZOL reagent. Each extracted RNA was used for the preparation of the miRNA sequencing library, which including 3′-adaptor ligation, 5′-adaptor ligation, cDNA synthesis and PCR amplification, approximately 150 bp PCR amplicons (corresponding to~22 nt miRNAs) size of products were selected. Constructed RIP-Seq and RNA-Seq libraries were controlled for quality and quantified using the BioAnalyzer 2100 system (Agilent Technologies, Inc., USA), and the libraries sequencing were performed on an illumina Hiseq instrument with 150 bp paired-end reads. Constructed miRNA-Seq libraries sequencing were denatured as single-stranded DNA molecules, captured on Illumina flow cells, amplified in situ as clusters and finally sequenced for 50 cycles on Illumina HiSeq Sequencer according to the manufacturer's instructions.
High-throughput sequencing for RIP-Seq, RNA-Seq and miRNA-Seq were all done by Cloud-Seq Biotech (Shanghai, China).
Analyses for High-throughput sequencing data. Paired-end reads were harvested from Illumina HiSeq 4000 sequencer, and were quality controlled by Q30. After 3′ adaptor-trimming and removing of low quality of reads by cut adapt software (version 1.9.3). For RIP-Seq, high quality clean of trimmed reads were aligned to the human reference genome (UCSC hg19) by bowtie2 software (version 2.2.4) with default parameters. The target binding regions of AGO2 were identified using MACS2 software (version 1.4.3). High-confidence binding regions of AGO2 were identified by stringent cutoff threshold, and then annotated with the latest UCSC RefSeq database to connect the peak information with the gene annotation. Fold enrichment of each mRNA transcript matched to different MACS identified motifs were summed. For RNA-Seq, high-quality clean reads were aligned to the reference genome (UCSC hg19) with hisat2 software (version 2.0.4). Then, guided by the Ensembl gtf gene annotation file, cuffdiff software (part of cufflinks) was used to get the gene-level FPKM (Fragments per kilobase of exon per million fragments mapped) as the expression profiles of mRNA, and fold change and pvalue were calculated based on FPKM, differentially expressed mRNA were identified.
For miRNA-Seq, raw data were generated after sequencing, image analysis, base calling and quality filtering on Illumina sequencer and finally quality controlled by Q30. The adaptor sequences were trimmed and the adaptor-trimmed-reads (>= 15 nt) were left by cut adapt software (version 1.9.2). Then the trimmed reads from all samples were pooled, and miRDeep2 software (version 2.0.0.5) was used to predict novel miRNAs. The trimmed reads were aligned to the merged human pre-miRNA databases (known pre-miRNA from miRBase plus the newly predicted pre-miRNAs) using Novoalign software (version 3.02.12) with at most one mismatch. The numbers of mature miRNA mapped tags were defined as the raw expression levels of that miRNA. The read counts were normalized by TPM (tag counts per million aligned miRNAs) approach. Differentially expressed miRNAs between two samples were filtered through Fold change. miRNA targets were performed by popular miRNA target prediction by using TargetMiner.