Nonsense‐mediated decay factor SMG7 sensitizes cells to TNFα‐induced apoptosis via CYLD tumor suppressor and the noncoding oncogene Pvt1

Yang et al. found that mutation of the nonsense‐mediated decay factor SMG7 enhanced resistance to tumor necrosis factor (TNF)‐induced apoptosis via two oncogenic pathways. Downregulation of CYLD and overexpression of the long noncoding RNA Pvt1 converge to promote NF‐κB activation. The results indicate a composite tumor suppressor role for SMG7 in response to TNF.

Nonsense-mediated decay (NMD) proteins are responsible for the surveillance and degradation of aberrant RNAs. Suppressor with morphogenetic effect on genitalia 7 (SMG7) is an NMD complex protein and a regulator of tumor necrosis factor (TNF)-induced extrinsic apoptosis; however, this unique function has not been explored in detail. In this study, we show that loss of Smg7 leads to unrestricted expression of long noncoding RNAs (lncRNAs) in addition to NMD targets. Functional analysis of Smg7 À/À cells showed downregulation of the tumor suppressor cylindromatosis (CYLD) and diminished caspase activity, thereby switching cells to nuclear factor-jB (NF-jB)-mediated protection. This positive relationship between SMG7 and CYLD was found to be widely conserved in human cancer cell lines and renal carcinoma samples from The Cancer Genome Atlas. In addition to CYLD suppression, upregulation of lncRNAs Pvt1 and Adap-t33 rendered cells resistant to TNF, while pharmacologic inhibition of NF-jB in Pvt1-overexpressing TNF-resistant cells and Smg7-deficient spheroids re-established TNF-induced lethality. Thus, loss of SMG7 decouples regulation of two separate oncogenic factors with cumulative downstream effects on the NF-jB pathway. The data highlight a novel and specific regulation of oncogenic factors by SMG7 and pinpoint a composite tumor suppressor role in response to TNF.

Introduction
Dysregulated gene expression is a hallmark of progressive cancers. As a result of cancer-causing mutations, tumors upregulate unconventional transcripts such as noncoding, micro-RNA, and pseudogenes. Many of these transcripts are recognized by the presence of a premature termination codon (PTC) and are targeted for degradation. Although PTC-containing transcripts are tightly controlled in healthy untransformed cells, their dysregulation underlies many human diseases [1]. Up to 30% of all heritable cancers are thought to result from such nonsense-mutations [2].
The major process for degrading aberrant transcripts is known as nonsense-mediated decay (NMD). Surprisingly, more than 10% of mRNAs are susceptible to this phenomenon. These include PTCs and extended 3 0 UTRs, as well as many others with yet undefined signatures [3]. An emerging class of NMD targets includes long noncoding RNAs (lncRNAs). LncRNAs are polymerase II transcripts that lack an open reading frame but have diverse regulatory roles as protein cofactors, micro-RNA sponges, transcriptional enhancers, and antisense RNAs [4]. Due to this versatility, they play a wide-ranging role in human cancer [5,6]. It follows that onco-lncRNAs like ANRIL, GAS5, and MALAT1 are increasingly identified as tumor suppressors and oncogenes [4,7], whereas others such as HOTAIR serve as functional biomarkers [8,9].
SMG7 is an RNA surveillance factor that functions together with up-frameshift (UPF) factors to deadenylate and degrade target RNAs [10,11]. Several studies have highlighted SMG7 interaction with P53 as influencing cellular survival [12,13]. SMG7 has also been identified in a large cohort to be associated with prostate cancer [14]. Our previous work identified Smg7 in a whole-genome mutagenesis screen against TNFa, a pleiotropic cytokine that can induce extrinsic apoptosis [15]. TNFa can induce cytotoxicity in tumors [16] but also plays a central role in NF-jB activation and inflammation. Yet, the functional role of downstream targets of SMG7 with respect to TNFa and tumor biology is poorly understood.
In this study, we examined gene expression in Smg7 À/À cells and found that lncRNAs rather than PTC-containing transcripts were preferentially overexpressed, indicating that SMG7 uniquely targets this class of transcripts. Further evaluation of the TNFa pathway in these cells identified decreased CYLD tumor suppressor protein as the source of apoptosis resistance. CYLD is a negative regulator of NF-jB that acts at the pathway branchpoint between apoptosis and NF-jB activation. Accordingly, downregulation of CYLD in Smg7 À/À cells reduced caspase activity and promoted NF-jB-mediated survival, while Cyld overexpression and NF-jB pharmacological inhibition re-established TNFa sensitivity. Strikingly, CYLD and SMG7 expression showed a near-universal correlation in diverse human cancer cell lines and clear cell renal cell carcinoma patient survival.
We further examined noncoding RNAs as preferred degradation targets of SMG7. Overexpression of two lncRNAs, Pvt1 and Adapt33, showed robust protection against TNFa that increased further upon Smg7 knockdown. PVT1 is an oncogene identified in Burkitt's lymphoma [17], while Adapt33 is a stress-induced transcript upregulated in response to apoptotic stimuli [18]. Administration of TNFa to 3D spheroids produced widespread cell death in parental cells, while Smg7 À/À spheroids showed compaction with viability. Nevertheless, pharmacological sensitization of the NF-jB pathway in both cell lines suppressed CYLD-and Pvt1-mediated survival. Taken together, the direct relationship with CYLD and regulation of oncogenic lncRNA Pvt1 identify SMG7 as a key molecular switch for cell survival in response to TNFa.

Cell viability assays
Unless stated otherwise, 3 9 10 3 cells were counted by a ViCell cell counter (Beckman Coulter, Brea, CA, USA), seeded in 96-well plates, and treated with the respective compounds as indicated. For dose-response curves, serial dilutions of respective compounds were prepared in 100 µL medium and cells were added on top in 100 µL medium. Cell viability was assessed by the addition of Resazurin (Sigma, St. Louis, MO, USA) to final concentration of 50 µM, and fluorescence was measured 6 h later at 540 nm excitation/590 nm emission in a PerkinElmer Envision 2104 (PerkinElmer, Waltham, MA, USA) Multilabel plate reader. At least three wells per condition were averaged, and viability is presented as percentage relative to respective control. For growth analysis, 3 9 10 4 cells were seeded in 12well plates and counted by a ViCell cell counter (Beckman Coulter, Brea, CA, USA).

Generation of cell lines
Knockout of Smg7 in MF was previously described [15]. Overexpression and knockdown cell lines were generated using CRISPRa [19] and CRISPRi [20], respectively. Identification of transcriptional start sites was accomplished using FANTOM5 [21], and top guides (À200) were selected from the UCSC genome browser [22]. Cell pools overexpressing Adapt33 or Pvt1 were generated using the CRISPRa guides (Table S1, see Section 2.4. for details). For Cyld overexpression cells, the full-length murine Cyld was amplified from parental MF cell cDNA (Table S1) and cloned into the pIREShyg3 (Clontech, Mountain View, CA, USA) construct. After linearizing with SpeI, Cyld containing or empty control construct was transfected into MF Smg7 À/À cells. Cell pools were selected with 250 µgÁmL À1 Hygromycin B (Sigma), and CYLD protein overexpression was confirmed by western blot.

CRISPR activation screen
Thirty-one upregulated genes in the MF Smg7 À/À cells were selected from our RNAseq data. For each gene, three guides were designed (Sulf has 2 TSSs, Table S1). In total, 96 guides were cloned into lenti sgRNA(MS2)_ zeo backbone (gift from F. Zhang, Addgene #61427, Watertown, MA, USA) and a mini lentiviral library was used to infect MF-dCas9-MS2 screening cells as described [19]. Two days after infection, cells were selected with 200 µgÁmL À1 Zeocin (Life Technologies, Carlsbad, CA, USA) in a 10 cm dish for 5 days. Then, 3 9 10 5 cells were treated with 20 ngÁmL À1 TNFa (Thermo Fisher Scientific) in a 6-well plate for another 5 days with the replacement of fresh TNFa containing medium every 2 days. Following this, genomic DNA was isolated by phenol-chloroform extraction from TNFa-selected and untreated control pools. Triplicate guide sequences were amplified (Table S1) and sequenced on an IonTorrent platform, and deconvolution was performed as described [15].

qPCR
Total RNA was isolated using InviTrap Spin Cell RNA Mini Kit (Stratec, Birkenfeld, Germany), and 500 ng RNA was used for first-strand cDNA synthesis via random hexamer primer and AMV Reverse Transcriptase (NEB, Ipswich, MA, USA) following the manufacturer's instructions. The qPCR was carried out on a LightCycler 480 (Roche, Basel, Switzerland) using Power SYBR Green PCR Master Mix (Thermo Fisher Scientific). The program included an initial denaturation step of 95°C for 10 min and 40 cycles of denaturation at 95°C for 15 s and combined annealing and extension at 60°C for 1 min. A melting curve analysis confirmed desired single PCR amplicons. Relative expression levels compared to control conditions were calculated by the comparative 2 ÀDDCp method using Gapdh or Actin as a reference gene. qPCR primers are listed in Table S1.

Western blotting
Cells were lysed in lysis buffer (63 mM Tris-HCl, pH 6.8, 10% glycerol, 2% SDS, 2.5% DTT, and 19 protease inhibitor cocktail; Sigma) for 30 min on ice. DNA was shredded with a sonicator and was pelleted for 20 min at max speed centrifugation at 4°C. The supernatant was mixed with 49 Roti-Load (Roth) and run on a 6-18% gradient SDS/PAGE gel and transferred onto PVDF membranes (Roth, Karlsruhe, Germany) using electrophoretic semi-dry western blot transfer system (Phase, L€ ubeck, Germany). Membranes were blocked with 5% skim milk (Sigma) in TBS-T for 1 h at room temperature and then incubated with specific primary antibody (dilution 1 : 1000 in 2.5% BSA in TBS-T or 5% milk in TBS-T) overnight at 4°C. Membranes were washed three times for 5 min in TBS-T before the addition of HRP-coupled secondary antibody (1 : 2000 in 5% milk in TBS-T) for 1 h at room temperature. Chemiluminescence detection was conducted using ECL prime western blotting detection reagent (GE Healthcare, Chicago, Illinois, USA) according to the manufacturer's instructions. Experiments were repeated independently at least two times, and antibody specificity was confirmed by comparing bands to molecular weight of the respective target protein. Antibodies used in this paper are listed in Table S1.

Lentiviral transduction
Third-generation lentiviruses were made using pHCMV-EcoEnv (Addgene #15802), pRSV-Rev (Addgene #12253), and pMDLg/pRRE (Addgene #12251) and the respective transfer vectors. Fresh 293T cells were seeded 1 day before with an appropriate number to reach 70% confluency. Plasmids were mixed with X-tremeGENE HP (Roche) at a ratio of 1 : 3 (DNA : reagent) in serumfree medium. The transfection complex was incubated at room temperature for 15 min and added dropwise onto 293T cells. Virus supernatant was collected after 72 h, filtered through a 0.45 µM Millex Syringe Filter (Merck Millipore, Darmstadt, Germany), and added to recipient cells. After infection for 48 h, cells were selected under the respective antibiotic to generate pools.

Luminescence assay
Caspase-8 and caspase-3/7 activity assays were performed using Caspase-Glo 8 and 3/7 Assay Systems (Promega, Madison, WI, USA) following the manufacturer's instructions. Briefly, 1.5 9 10 4 Smg7 À/À or parental MF cells per well were seeded 1 day before in a 96-well plate. On the second day, cells were exposed to 20 ngÁmL À1 TNFa (Thermo Fisher Scientific) for 8 h or left untreated. The medium was removed, and cells were washed twice with PBS. Fifty microlitre reagent and 50 µL PBS were added and incubated at room temperature with continuous shaking. After 20 min, luminescence was recorded on an Envision 2104 Multilabel plate reader (PerkinElmer). Empty wells served as blanks and were subtracted from all values.

3D spheroid culture
Five hundred parental MF cells and 1000 Smg7 À/À cells per well were seeded into Corning 4515 plates (Corning, NY, USA), grown for 4 days, and treated with 20 ngÁmL À1 TNFa and 10 µM SC-514 for 2 days, with replacement of medium after 1 day. Spheroids were stained with 1 µgÁmL À1 propidium iodide (PI) for 1 h and imaged in a PerkinElmer Operetta High Content Imaging System using brightfield (BF) and 535/30 nm excitation and 595/70 nm emission channel.
2.11. Quantitative mass spectrometry 1 9 10 7 Smg7 À/À and parental MF cells per replicate (n = 3) were lysed in 8 M urea in 0.1 M Tris/HCl pH 8.5 using a Precellys homogenizator (Bertin Technologies, Montigny-le-Bretonneux, France). Equal amounts were proteolysed using a modified FASP procedure [23]. Briefly, after reduction and alkylation using DTT and IAA, the proteins were centrifuged on Microcon centrifugal filters (Vivacon 500 30 kDa, Sartorius, Goettingen, Germany) and washed 39 with 8 M urea in 0.1 M Tris/HCl pH 8.5 and twice with 50 mM ammonium bicarbonate. The proteins were digested for 2 h at room temperature using 0.5 µg Lys-C (Wako Chemicals, Neuss, Germany) and for 16 h at 37°C using 1 µg trypsin (Promega). Peptides were eluted by centrifugation (10 min at 14 000 g), acidified with 0.5% TFA, and stored at À20°C. Approximately 0.5 µg of peptides per sample was measured on a Q-Exactive HF mass spectrometer online coupled to an Ultimate 3000 nano-RSLC (Thermo Scientific, Waltham, MA, USA) in data-independent acquisition (DIA) mode as described previously [24]. The recorded raw files were analyzed using the SPEC-TRONAUT PULSAR software [25] with a peptide and protein identification false discovery rate setting of < 1%, using an in-house mouse spectral meta library which was generated using PROTEOME DISCOVERER 2.1 (Thermo Scientific), the Byonic search engine (Protein Metrics, Cupertino, CA, USA), and the SwissProt Mouse database (release 2016_02). Quantification was based on MS2 area levels of all unique peptides per protein fulfilling the percentile 0.3 setting. Normalized protein quantifications were exported and used for calculations of fold changes and significance values.

RNA sequencing and analysis
Independent replicates of Smg7 À/À and parental controls (n = 3) were harvested, and RNA extracted using the RNeasy Mini Kit (Qiagen, Hilden, Germany) and ribosomal RNA (rRNA) subtractive sequencing was performed as 150 bp paired-end runs on Illumina HiSeq 4000 platform. RNAseq analysis was performed on Galaxy with alignment against mouse genome version mm10.gtf using Salmon transcript quantification [26].
Gencode vM17 transcripts were used as reference transcriptome with transcript-to-gene mapping. Sequence data were visualized with Broad IGV. Biotype assignments were retrieved from Ensembl/Gencode through the Biomart function. Transcript counts below 1 were excluded from further analysis. Significance between Smg7 À/À and parental cells for each biotype was determined using DESEQ2 [27]. Differential gene expression analysis was performed on rRNA-depleted samples with TOPHAT and CUFFLINKS [28] using default settings on Galaxy and mouse genome version mm10. TOPHAT alignment to the file gencode.vM12.lncRNA_transcripts.fa was performed to determine percent transcript mapping to lncRNA. Cuffdiff was used to calculate FPKM levels for each gene with Cuffdiff gene level determinations with the default settings.

Gene set enrichment analysis
Gene set enrichment analysis (GSEA) was performed using top upregulated and downregulated (n = 100 each) genes as described [29].

Association between SMG7 and CYLD and patient survival
Kidney renal clear cell carcinoma (KIRC) transcriptome data and clinical information were downloaded via The Cancer Genome Atlas [32] on December 15, 2019. The correlations between gene expression levels were calculated by Pearson's test in R version 3.53 (https://www.r-project.org/). A Kaplan-Meier plot was applied to identify the association between SMG7 expression and patient survival. Log-rank test and univariate Cox regression were used to determine significance. Data were visualized using GGPLOT2 [33]. Briefly, all tumor samples were centered into nearest neighbor values by their expression level of CYLD. Each dot in the plot represents averaged SMG7 expression level of 40 unique tumor samples.

Statistics/data analysis
General data visualization and statistics were performed in GRAPHPAD PRISM. If not stated otherwise, Student's two-tailed, unpaired t-test against respective control conditions was used to determine significance.

Apoptosis-resistant Smg7 À/À cells upregulate noncoding RNAs
Depending on the cellular context, TNFa can either induce cell death or activate the NF-jB pathway. We have shown that Smg7 ablation in immortalized mouse fibroblasts (MF) protects against TNFa-induced apoptosis but not chemically induced cell death [15]. We reanalyzed Smg7 À/À MF cells for TNFa sensitivity and observed Z-VAD-FMK (zVAD) rescue, demonstrating caspase involvement (Fig. 1A). Independent knockdown of Smg7 by CRISPR interference [20] in NIH 3T3 cells (Smg7KD) recapitulated the TNFa-resistant phenotype (Fig. S1A). In both instances, a consistent fraction of cells was insensitive to TNFa-induced apoptosis, likely due to impeded signaling in cellular aggregates.
Next, we tested whether Smg7 loss affects TNFa signaling exclusively or interferes with other forms of cell death. We treated MF cells with TNF-related apoptosis-inducing ligand (TRAIL), TNF-related weak inducer of apoptosis (TWEAK), and lipopolysaccharide (LPS), but observed insensitivity to cell death in parental and Smg7 À/À cells, irrespective of cycloheximide or IFNc sensitization (Fig. S1B). A panel of chemotherapeutics showed minor protection in Smg7 À/À cells against doxorubicin and staurosporine and partial zVAD-independent sensitivity to paclitaxel (Fig. S1C). This shows increased viability due to Smg7 mutation is largely stimulus-dependent and primarily affects TNFa signaling.
Classical targets of NMD include PTC-containing transcripts and lncRNAs. We analyzed hallmark lncRNAs for downstream effects due to the loss of Smg7 by quantitative PCR (qPCR). Three transcripts, Gas5, Anril, and Hotair, were elevated~2-to 6-fold, while Malat1 was decreased (Fig. 1B). We suspected that these relatively minor changes were due to functional redundancy in the NMD complex or as compensation for SMG7. However, analysis of a panel of NMD effectors showed only minor increases in Smg5 and Upf3b transcripts, suggesting negligible compensation (Fig. 1C). We therefore examined whether ablation of these same NMD factors could also protect against TNFa using data from our pan-genomic mutagenesis screen in the same cells [15]. Only basal guides directed against Smg7 and Upf2 were found to be enriched. This suggests that resistance to TNFa-induced apoptosis is mediated by specialized function of SMG7 rather than by global disruption of NMD.
We therefore investigated whether particular classes of transcripts are affected inSmg7 À/À cells using RNA sequencing. We grouped resulting transcripts by biotype and observed that prototypical NMD targets are most enriched in the top 100 upregulated transcripts. Notably, protein-coding genes, miRNAs, and processed transcripts (consisting of lncRNAs and antisense RNAs) comprise 45% of the top upregulated transcripts (Fig. 1E). The majority (78%) of downregulated transcripts, in contrast, are almost exclusively protein-coding genes.
Strikingly, global differential expression analysis revealed long intergenic noncoding RNAs [lincRNA; Fig. 1F; P(adj) % 0.0000] and antisense transcripts [P (adj) = 1.84E-67] to be most significantly upregulated, while protein-coding genes showed only a nominal decrease (Fig. 1F). Separate subclasses of lncRNAs, including small nuclear RNAs (snRNA), sense intronic, and nonstop decay transcripts, were similarly elevated. Global NMD targets, in contrast, only showed a limited increase. As expected, the miRNA class of transcripts was decreased [P(adj) = 5.65E-22], consistent with a lncRNA role as molecular sponges [34]. Gene set enrichment analysis confirmed cancer pathway features and downregulation of apoptosis genes (Fig. S2). A separate proteomic analysis of Smg7 À/À cells also showed accumulation of cytokine response factors (such as Mgst3) and caspase-1 (Fig. S3). Proteins that negatively regulate viruses also featured, possibly as a consequence of expressing unconventional transcripts. Pseudogenes mRNAs contain PTCs that target them for NMD; therefore, these would be expected to be elevated [35]. However, pseudogene transcripts were largely reduced in Smg7 À/À cells (Fig. 1F). Together, these data suggest that SMG7 preferentially targets processed transcripts such as lncRNAs over PTC-containing RNAs. Additionally, as discussed in Section 3.4., potent regulation of individual transcripts argues for selective target RNA regulation by SMG7.

Smg7 À/À cells show decreased caspase activity and CYLD levels
Extrinsic apoptosis is executed by caspases in the death-inducing signaling complex (DISC). To determine where in the TNFa signaling cascade SMG7 acts, we first examined caspase-8 and -3 activity. Using a luminescence assay, we observed that caspase-8, and more significantly, caspase-3 activity was strongly  decreased in Smg7 À/À cells stimulated with TNFa ( Fig. 2A). Diminished activity was corroborated in a TNFa time course by western blot showing a reduction of the pro-caspase form and accumulation of the mature cleaved enzyme in parental cells. These were largely absent inSmg7 À/À cells (Fig. 2B).
In light of reduced caspase activity, we inferred that SMG7 must functionally act on or upstream of these enzymes. We examined TNFa signaling receptors TNFR1 and -2, but detected no changes in protein expression to account for apoptosis inhibition (Fig. 2C). Receptor-associated complex I factors TRAF2 and RIPK1 expression were similarly unaffected; however, CYLD deubiquitinase showed a striking decrease in basal levels in Smg7 À/À cells compared to parental cells (Fig. 2C). CYLD acts at the interface between cell death and survival pathways. Therefore, we tested and confirmed a strong increase in phosphorylated-IjB and its associated degradation, indicating NF-jB pathway activation, in Smg7 À/À cells.
Activation of Jun-kinase (phospho-JNK) in parental cells [36] and other effectors of the mitochondrial cell death pathway P53 and BID remained unchanged compared to parental cells. Together, these data confirm that SMG7 acts upstream of JNK/caspases but downstream of surface receptors to sensitize cells to TNFa.
CYLD is a tumor suppressor that deubiquitinates K63 chains directly downstream of TNFa receptors. When CYLD is diminished, the NF-jB survival pathway is activated [37]. We therefore tested NF-jB activation in a short interval time course and observed that the classical activation-feedback inhibition-activation response was unaltered in Smg7 À/À cells following TNFa administration (Fig. 2E). This indicates that functional TNFa signaling is enabled through its receptors. Moreover, A20, a transcriptional target and NF-jB effector, was strongly increased in Smg7 À/À cells. Therefore, we tested whether TNFa-resistance in Smg7 À/À cells is dependent on NF-jB using pharmacological inhibitors SC-514 and MLN120B. Cell viability was strongly decreased upon treatment in the presence of TNFa, with Smg7 À/À cells requiring higher concentrations of both inhibitors to achieve equivalent levels of cell death (Fig. 2F). These results demonstrate that resistance to TNFa-induced apoptosis in Smg7 À/À cells is mediated by CYLD/NF-jB.

CYLD and SMG7 coordinate apoptosis sensitivity
We hypothesized that CYLD levels are qualified by SMG7 and evaluated this relationship in other cell types. Forced knockdown of SMG7 by siRNA (siSmg7) in NIH 3T3 and MCF-7 breast cancer cells restored viability upon TNFa treatment (Fig. 3A). In a striking similarity to MF cells, SMG7 knockdown in both cell lines resulted in CYLD downregulation and TNFa resistance, suggesting conservation of this relationship. We therefore tested whether ectopic Cyld overexpression (Cyld OE) in Smg7 À/À cells could override TNFa resistance and observed a partial resensitization compared to empty vector control (Fig. 3B). Correspondingly, Cyld siRNA knockdown (siCyld) partially increased resistance of parental MF cells to TNFa (Fig. 3B). Together, these results indicate the SMG7/CYLD relationship is conserved and that direct manipulation of CYLD expression influences apoptosis sensitivity.
We chose to examine whether this relationship extended further. We therefore compared CYLD and SMG7 expression levels in 1164 human cancer cell lines [31] and found a comprehensive positive correlation as well as marked Pearson correlations for soft tissue (R = 0.65; Fig. 3C), multiple myeloma (R= 0.58), rhabdoid (R = 0.605), upper aerodigestive (R = 0.397), and gastric cancers (R = 0.407). A moderate correlation (R = 0.41) was observed in primary renal cell carcinoma samples [32]; however, RCC cells have established susceptibility to TNF family-induced apoptosis through the TNF receptors [38,39] and plasma TNFa levels are a prognostic indicator for RCC. Higher TNFa levels may therefore indicate refractory, or NF-jB activating, tumors. Consistent with this, Kaplan-Meier analysis of 516 KIRC-TCGA patients revealed a strong correlation between elevated SMG7 expression and survival [P(HR) = 0.00022; Fig. 3D].
Many lincRNAs have not yet been functionally characterized. We wondered whether top enriched genes, including protein-coding genes, could functionally recapitulate Smg7 À/À resistance to TNFa. A CRISPR activation [19] library was designed targeting the promoters of top upregulated genes ( Fig. 4A and Fig. S4A). We selected only genes with a clear transcription start site (TSS) and used a tiling function within À200 bp of the TSS. Parental MF cells were transduced with the pooled CRISPRa mini-library and treated with TNFa for 5 days to induce extrinsic apoptosis. Genomic DNA from surviving cells was harvested for deep sequencing, and guides were deconvoluted utilizing ENCORE software [15]. After multiple hypothesis correction, only two lncRNAs were found to significantly protect against TNFa challenge: the oncogene plasmacytoma variant translocation 1 (Pvt1; P = 0.000865, FDR = 0.0031) and 5430416N02Rik, also known as Adapt33 (P < 0.0001, FDR < 0.0001; Fig. 4B). The decoy TNF receptor Tnfrsf11b also displayed protection, likely via interference at TNFRSF1A/B signaling receptors. However, given downstream NF-jB activation in Smg7 À/À cells (Fig. 2C,E), we surmised that TNF receptor signaling at the membrane is not substantially impeded. Thus, we focused our investigation on the two lncRNAs.
Adapt33is a stress-induced lncRNA induced by hydrogen peroxide or staurosporine [18]. We examined lncRNA expression by qPCR in Smg7 À/À cells and observed a 7-and 10-fold increase in Pvt1 and Adap-t33 transcripts, respectively, that increased upon TNFa treatment (Fig. 4C). We therefore generated Pvt1 and Adapt33 overexpression cell lines (OE; Fig. S4C) and evaluated resistance to TNFa. Compared to control cells containing an empty vector, Pvt1 OE cells were significantly more viable at 5 ngÁmL À1 TNFa (P = 0.0001; Fig. 4D). Adapt33 OE also displayed  protection, albeit at lower levels (P = 0.0013). We reasoned that incomplete protection was due to persistent SMG7 surveillance of overexpressed transcripts. Inclusion of siRNA directed against Smg7 (siSmg7; Fig. S4D) substantially potentiated Pvt1 OE and Adapt33 OE resistance to TNFa (Fig. 4D). However, we determined that Pvt1 OE and Adapt33 OE, while individually protecting against TNFa, do not depress CYLD levels detectably (Fig. 4D). Caspase-8 and -3 cleavage in both cell lines was comparable to parental cells (Fig. 4E). However, a synergistic sensitization to TNFa was observed upon Pvt1 knockdown in Cyld overexpression cells (Fig. 4F). Thus, protection afforded by these lncRNAs likely mechanistically complements the SMG7-CYLD relationship. Nevertheless, the NF-jB inhibitor SC-514 sensitized Pvt1 OE cells to TNFa (Fig. 4G), suggesting that Pvt1 OE resistance also relies on the activation of NF-jB pathway.

Pharmacological sensitization of Smg7 À/À cells to TNFa in 3D spheroid model
TNFa has pluripotent effects on tumorigenesis and cancer progression and an autocrine function in the tumor microenvironment. To mimic in vivo tissue and cellular communication, we examined the effect of Smg7 deletion and TNFa in a 3D spheroid model [40] ( Fig. 5A). We observed that Smg7 À/À cells have a growth disadvantage compared to parental cells (Fig. 5B). Seeding 500 parental and 1000 Smg7 À/À cells compensated for this difference yielding comparable spheroids. TNFa sensitivity of parental cells indicated by propidium iodide (PI) was preserved in this environment, while synthetic lethality with 10 µM SC-514 IKK inhibitor was increased (Fig. 5A). TNFa treatment of Smg7 À/À cells resulted in compacted, sharply defined spheroids, but did not affect viability. Pharmacological sensitization of Smg7 À/À cells with SC-514 nevertheless produced comparable cell death to TNFa treatment of parental cells.

Discussion
TNFa plays dynamic but often paradoxical roles in human malignancies [41]. Systemic TNFa can activate inflammation and pro-survival pathways via NF-jB, while higher local concentrations, that is, monocyte presentation or natural killer induced cytotoxicity, can induce apoptosis in tumors [42,43]. Fascinatingly, these contradictory outcomes are both triggered via by TNFa binding to its cognate receptors TNFRSF1/2 but ultimately decided by cellular factors directly downstream. The abundance of these factors in localized complexes is crucial in this decision. In particular, TNFa cytotoxicity is strongly influenced by the NF-jB pathway and associated factors [44].
In this study, we determined that SMG7 dictates levels of CYLD deubiquitinase and that loss of SMG7 leads to lower CYLD levels and increased cell survival upon TNFa challenge. The primary consequence of CYLD downregulation is the activation of the NF-jB pathway. This implies that cancers associated with constitutive NF-jB activation are more likely to be susceptible to SMG7 levels. Indeed, NF-jB therapeutics for prostate cancer, for which SMG7 was found to be a novel risk factor [14], shows promise in modulating tumorigenesis and progression [45]. Paracrine application of TNFa is known to induce apoptosis and cell cycle arrest in prostate cancer cells [46]. A direct relationship was also observed by CYLD ablation in testicular cells, resulting in NF-jB activation and aberrant expression of anti-apoptotic genes [47]. These results are entirely consistent with the phenotype of Smg7 À/À cells reported here, including resistance to TNFa and pharmacological sensitization to the combination of NF-jB inhibitors and TNFa in monolayers and 3D spheroids.
While the mechanism of SMG7 activity directly on specific target RNAs remains to be fully established, the present results demonstrate that, in addition to NMD transcripts, SMG7 targets a unique subset of lncRNAs. Whether SMG7 targets these transcripts on its own or in concert with other NMD proteins remains to be determined. SMG7 alone may contain the capacity to degrade RNA transcripts via interaction with CNOT8 [11] and can also degrade 3 0 UTR length-dependent mRNA via UPF1/SMG7-dependent miRNA-mediated mRNA decay pathway [48]. Screening results of other NMD factors, in contrast to Smg7, did not reveal protection against TNFa (Fig. 1C). Thus, our data support a model of specialized transcript degradation over a generalized contribution to NMD, with individual lncRNAs and mRNAs showing profound differences.
One of the most highly upregulated transcripts in Smg7 À/À cells is Pvt1. Pvt1 overexpression, as well as the lncRNA Adapt33, was sufficient to protect MF cells from TNFa-induced apoptosis. The relationship between SMG7, Pvt1, and CYLD is mechanistically complex (Fig. 5C). PVT1 was originally identified in Burkitt's lymphoma. It is located chromosomally adjacent to and regulates MYC [49], which in turn is responsible for the majority of Burkitt's lymphoma. SMG7 is highly expressed in cells derived from Burkitt's lymphoma [30] ] Smg7-/parental PVT1 [49]. PVT1 lncRNA does not directly regulate CYLD, but CYLD is implicated in controlling MYC via JNK [50], providing a hint to this relationship.
Adapt33 was reported as an oxidant-inducible and apoptosis-associated RNA [51]; however, its function has not yet been demonstrated. Here, we report that Adapt33 mRNA level is upregulated in Smg7 À/À MF cells and overexpression of Adapt33 by CRISPRa system can give protection to MF cells against TNFa-induced apoptosis. Adapt33 mRNA is also increased in MF cells after TNFa treatment in our study. The underlying anti-apoptotic mechanism of Adapt33, however, is still uncharacterized.
Long noncoding RNAs can impact multiple pathways simultaneously. As testament to this complexity, PVT1-encoded lncRNA or derived microRNAs have oncogenic functions [52]. Additionally pointing to miRNAs, CYLD transcripts have been shown to be targeted by miR-362-5p and miR-19 [53,54]. Thus, the possibility exists that other lncRNAs upregulated in Smg7 À/À cells may be processed to miRNAs that target Cyld. Indeed, a precedent already exists as miR-20a induces cisplatin resistance in human gastric cancer cells by targeting CYLD [55]. miRNAs from the PVT1 locus have also been suggested to have a regulatory function [52]. Similarly, HOTAIR overexpression maintains NF-jB expression and platinum chemoresistance [56], although a link via CYLD has not been explored. Certainly, the role of such multifactorial transcripts in cancer is being further dissected through large-scale genomic studies [57].

Conclusions
A major open question in oncology is how lncRNAs influence cell death pathways. Here, we show that loss of the NMD factor SMG7 in several cell types uniquely protects against TNFa-induced apoptosis. Strikingly, this phenotype is mediated by two different oncogenic factors. CYLD is positively correlated with SMG7, while overexpression of Pvt1 increases cell viability toward TNFa treatment. Taken together, our findings support a novel role of SMG7 in TNFa-induced apoptosis by regulating Pvt1 and the tumor suppressor CYLD and suggest a comprehensive role for regulation of NF-jB by an NMD factor.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Fig. S1. Effects of Smg7 ablation on cell death inducers. Fig. S2. Gene Set Enrichment Analysis in Smg7 À/À cells. Fig. S3. Quantitative mass spectrometry analysis of Smg7 À/À compared to parental MF cells. Fig. S4. Transcriptional changes in Smg7 À/À cells. Table S1. Primers, guides and antibodies used in this study. Table S2. Quantitative mass spectrometry data.