SMG1:SMG8:SMG9-complex integrity supports efficient execution of nonsense-mediated mRNA decay

Abstract

Nonsense-mediated mRNA decay (NMD) is a translation-dependent mRNA turnover pathway, which degrades transcripts containing premature termination codons. NMD activation depends on phosphorylation of the RNA helicase UPF1 by the SMG1 kinase, which acts in a complex with SMG8 and SMG9. Structural and biochemical studies have implicated SMG8 and SMG9 as regulators of SMG1 activity, but their contributions to NMD in human cells remain incompletely defined. Here, we systematically dissect the roles of SMG8 and SMG9 in NMD using genetic and pharmacological perturbations in multiple human cell lines. Deletion of the kinase inhibitory domain (KID) of SMG8 did not affect UPF1 phosphorylation or NMD efficiency, demonstrating that this domain is dispensable in vivo. Complete loss of SMG8 or SMG9 resulted in only modest NMD impairment and was accompanied by moderately increased UPF1 phosphorylation. However, SMG8- or SMG9-deficient cells exhibited pronounced hypersensitivity to partial pharmacological inhibition of SMG1, leading to synergistic, transcriptome-wide stabilization of NMD targets. These effects were reproducible across different cellular contexts, underscoring a general regulatory role for SMG8 and SMG9. Together, our results establish SMG8 and SMG9 as nonessential modulators that safeguard the efficiency and perturbation tolerance of the NMD pathway in human cells.

Graphical Abstract

Graphical Abstract.

Introduction

Alongside other co-translational quality control mechanisms, nonsense-mediated mRNA decay (NMD) detects and eliminates defective or undesired mRNAs [1–3]. The primary role of NMD is to identify messenger RNAs (mRNAs) carrying premature translation termination codons (PTCs) caused by mutations, alternative splicing, and other means [4]. The activity of NMD prevents the production of C-terminally truncated and possibly toxic proteins [5, 6]. Beyond its role in quality control, NMD also plays a role in regulating gene expression, thereby directly or indirectly affecting approximately 20%–40% of genes [7]. In many cases, NMD is activated by the exon junction complex (EJC). This multi-protein, RNA-binding complex is deposited 20–24 nucleotides (nts) upstream of exon–exon junctions by the spliceosome and remains bound to the mRNA during its export to the cytoplasm [8]. Translation termination at a PTC differs from that at a normal termination codon when it is located sufficiently upstream of an exon–exon junction bound by an EJC, whose presence can activate NMD [9–11]. The EJC is bridged by UPF3A/B and UPF2 to UPF1, the central NMD factor [12]. Subsequently, the unstructured N- and C-terminal tails of UPF1 are phosphorylated at [S/T]Q motifs by the SMG1:SMG8:SMG9 complex, which consists of the kinase SMG1 and its regulators SMG8 and SMG9 [13–17]. UPF1 harbours 28 [S/T]Q motifs, of which 19 are evolutionarily conserved [18]. While no single phosphorylation site appears to be indispensable for NMD, the synergistic effect of phosphorylation at multiple sites contributes to the degradation of NMD-targeted RNAs, with each site having a varying degree of importance [18]. The heterodimer of SMG5:SMG7 selectively binds to phosphorylated UPF1, whereas the endonuclease SMG6 interacts both in a phosphorylation-dependent and -independent manner with UPF1 [18–22]. Nevertheless, the SMG5:SMG7 complex is required for endonucleolytic cleavage of NMD substrates near the PTC [7, 22–25]. This activity is mediated by a composite PIN domain formed by the inactive PIN domain of SMG5 together with the weakly active PIN domain of SMG6 [26, 27, 28].

In metazoans, the conserved SMG1:SMG8:SMG9 complex plays an important role in ensuring the precise execution of the NMD processes. SMG1 belongs to the phosphatidylinositol-3-kinase-related kinase (PIKK) family, and phosphorylates serine or threonine residues [14]. It prefers a glutamine residue at position + 1 and leucine residue at -1 position for efficient phosphorylation [13, 14, 29]. Structurally, SMG1 comprises of a catalytically active C-terminal head and an N-terminal arm including the N-HEAT domain forming an alpha-solenoid [30]. The so-called insertion domain of SMG1 functions as PIKK-regulatory domain (PRD) and its removal leads to hyperactivation of the kinase [31, 32]. Functionally, the SMG1 insertion domain inhibits substrate binding and blocks the access to the active site [33].

SMG8 is composed of an N-terminal G-like domain followed by a C-terminal kinase inhibitory domain (KID), while SMG9 features a C-terminal G-domain. SMG8 and SMG9 form an unusual heterodimer with SMG9’s G-domain and SMG8’s G-like domain interaction mirroring that of dimeric GTPases [34]. On the side opposite to SMG8, the G-domain of SMG9 interacts with SMG1 via the N-HEAT domain and the alpha-solenoid [16, 29, 30, 34]. SMG8 interacts with SMG1 only via the alpha-solenoid, enabling SMG9 to control the activity of SMG1 indirectly via integration of SMG8 into the complex [16, 29, 30, 34]. Previous studies have shown that the removal of SMG8 or the deletion of its KID resulted in increased SMG1 activity in vitro, suggesting a regulatory role of SMG8 in inhibiting SMG1 via its KID [16, 30–33, 35]. Mechanistically, the KID stabilizes SMG1 in its autoinhibited state, offering insight into how SMG8 regulates SMG1 activity at a molecular level [33].

The existing knowledge about SMG8 and SMG9 is largely derived from analyzing recombinant proteins in vitro, leaving a gap in understanding their contributions to NMD regulation within intact cells. In an endeavour to bridge this knowledge gap, our initial hypothesis suggested that deleting the KID of endogenous SMG8 would enhance SMG1 activity, resulting in UPF1 hyperphosphorylation. However, we detected no changes in steady-state phosphorylation, which prompted us to generate SMG8- and SMG9-knock-out (KO) cells via CRISPR–Cas9, resulting in mild NMD impairment and moderately increased UPF1 phosphorylation. Treatment of SMG8 and SMG9 KO cells with SMG1 inhibitor (SMG1i) resulted in severe NMD impairment, underscoring the functional regulation of SMG1 by SMG8 and SMG9. We analyzed the transcriptome-wide effects of SMG1i via RNA-Seq and detected concentration and KO-dependent upregulation of NMD-annotated transcripts. Analysis of the interactome of immunoprecipitated endogenous UPF1 suggests that SMG8 and SMG9 KOs, as well as SMG1i treatment, alter the composition of NMD complexes, reflecting functional inhibition of NMD at different stages of NMD machinery assembly. Collectively, these results provide an extensive in vivo characterization of the SMG1:SMG8:SMG9 complex, whose integrity is required for the efficient and perturbation-tolerant execution of NMD.

Materials and Methods

Cell lines

HCT116 (human, male, colorectal carcinoma, epithelial; ATCC, cat. no. CCL-247; RRID:CVCL_0291), Flp-In-T-REx-293 (human, female, embryonic kidney, epithelial; Thermo Fisher Scientific, cat. no. R78007; RRID:CVCL_U427) and U2OS Flp-In-T-REx (human, female; gift from Stephanie Panier) were cultivated in high glucose, GlutaMAX DMEM (Gibco) supplemented with 9% fetal bovine serum (Gibco) and 1× Penicillin–Streptomycin (Gibco). The cells were cultured at 37 °C and 5% CO₂ in a humidified incubator. The generation of knock-in/knock-out and stable cell lines is described below. All cell lines are summarized in Supplementary Table S1. Cells were maintained mycoplasma-free and regularly checked for contaminations by polymerase chain reaction (PCR) or using the Mycoplasmacheck service from Eurofins Genomics.

Generation of knock-out and knock-in cells using CRISPaint or CRIS-PITCh system

The SMG8 and SMG9 knock-out HCT116 cells were generated via the CRISPaint system [36]. The single guide RNA (sgRNA) sequence for SMG8 delKID was 5′-CTATTGTGATATAGCACAGG-3′, for SMG8 KO 5′-AGCTTGCGAGACCTTCTAAT-3′ and for SMG9 KO 5′-TGCGCCACCCAAGGGGGAGA-3′. 2.5 × 10⁵ cells per sgRNA were seeded in six-well plate. One day after seeding, 2000 ng universal donor (pCRISPaint-myc-PuroR, Addgene plasmid # 80961; pCRISPaint-TagGFP2-PuroR, Addgene plasmid # 80970; both vectors were a gift from Veit Hornung), 1000 ng frame selector (pCAS9-mCherry-Frame + 0/+1/+2, Addgene plasmid # 66 939/# 66 940/# 66 941; all three plasmids were a gift from Veit Hornung), and 1000 ng target selector (pX330-SMG8 delKID-R57; pX330-SMG8-IDT-AA; pX330-SMG9-chop94) were transfected using Lipofectamine 2000 (Thermo Fisher Scientific) according to the manufacturer’s protocol. 2 days after transfection the cells were transferred to 10 cm dishes and 4–5 days after transfection cells were selected with 0.75–1.0 μg/ml Puromycin (InvivoGen). All CRISPaint plasmids are summarized in Supplementary Table S1.

The SMG8 and SMG9 knock-out U2OS and HEK293 cells were generated via the CRIS-PITCh v2 system [37]. The plasmid pX330-PITCh_SMG8 is based on pX330-BbsI-PITCh (Addgene plasmid # 127875; was a gift from Peter Kaiser) and contains the SMG8 sgRNA (5′- AGCTTGCGAGACCTTCTAAT -3′); the plasmid pX330-BbsI-PITCh-SMG9 contains the SMG9 sgRNA (5′-CGCTCTATCCCATAGAGTCC-3′), respectively. The donor plasmid pCRIS-PITChv2-SMG8_PurR_KO is based on pCRIS-PITChv2-dTAG-Puro (BRD4) (Addgene plasmid # 91796; was a gift from James Bradner & Behnam Nabet) and contains two 40 bp-long microhomologies (5′-CTGCACTATGGCTGGTCCCGTGAGCTTGCGAGA CCTTCTA-3′ and 5′-GGAGCATCAGCCTGGATGGGCTCT GAAAGTCCCGGAGGGT-3′) flanking a P2A signal, a puromycin resistance gene, a stop codon, and a poly(A) signal. The donor plasmid pCRIS-PITChv2-SMG9-KO contains a 5′ microhomology (5′-ATAGGTAACCATGTCTGAGTCTGGACACAGTCAGC CTGGA-3′) and a T2A signal, a puromycin resistance gene, a stop codon, and a poly(A) signal.

For SMG8 KOs, 2.5 × 10⁵ cells were in seeded in a six-well plate and after 24 h 1000 ng pX330-PITCh_SMG8 and 500 ng donor plasmid pCRIS-PITChv2-SMG8_PurR_KO were transfected using the calcium phosphate method. For SMG9 KOs, 2.5 × 10⁵ cells were seeded in a six-well plate and after 24 h 1000 ng pX330-PITCh_SMG9 and 500 ng donor plasmid pCRIS-PITChv2-SMG9_PurR_KO were transfected using the calcium phosphate method. Two days after transfection the cells were transferred to 10 cm dish and 3 days after transfection cells were selected with 1 μg/ml puromycin (InvivoGen).

UPF1 was endogenously FLAG-tagged via the CRIS-PITCh v2 system [37]. The plasmid pX330-BbsI-PITCh-UPF1-N encodes the UPF1-specific sgRNA (5′-CCCGTACGCCTCCACGCTCA-3′). The donor plasmid pCRIS-PITChv2-HygR-FLAG contains two 40 bp-long N-terminal UPF1 microhomologies (5′-GCAGCGCGGAACCGGCCCGAGGGCCCTACC CGGAGGCACC-3′ and 5′-GAGCGTGGAGGCGTACGG GCCCAGCTCGCAGACTCTCACTT-3′) flanking a Hygromycin resistance gene, a T2A signal, the FLAG-tag, and a linker region. For transfection, 2.5 × 10⁵ cells were in seeded in a six-well plate and after 24 h 1000 ng pX330-BbsI-PITCh-UPF1-N and 500 ng donor plasmid pCRIS-PITChv2-PurR-FLAG were transfected using Lipofectamine 2000 (Thermo Fisher Scientific) according to the manufacturer’s protocol. Two days after transfection, the cells were transferred to 10 cm dish and 4–5 days after transfection cells were selected with 100 μg/ml Hygromycin (InvivoGen). All CRIS-PITCh plasmids are summarized in Supplementary Table S1.

For both systems, cells were selected for 2–3 weeks with Puromycin or Hygromycin. Cell colonies originating from a single clone were isolated in 12-well plates and genomic DNA was extracted using QuickExtract DNA Extraction Solution (Lucigen) according to the manufacturer’s instruction. Correct insertion of the gene cassette was screened via genomic PCR and verified via Sanger sequencing (Eurofins Genomics). The primers for genomic PCR are listed in Supplementary Table S1.

DNA and RNA extraction

One day prior to DNA extraction, cells were seeded in a 48-well plate. To extract DNA, 50 μl QuickExtract DNA Extraction Solution (Lucigen) was used following the manufacturer’s instructions.

For RNA extraction, cells were dissolved in 1 ml in-house prepared TRI reagent [38] per six-well and RNA was extracted following instructions of peqGOLD TriFast (VWR Peqlab; v0815_e). Following changes were made: instead of 200 μl of chloroform, 150 μl of 1-bromo-3-chloropropane (Sigma–Aldrich) was used. RNA was resuspended in 20 μl of RNase-free water.

Western blot analysis

For SDS–polyacrylamide gel electrophoresis and western blot analysis protein samples were harvested with RIPA buffer (50 mM Tris–HCl pH 8.0, 0.1% SDS, 150 mM NaCl, 1% IGEPAL CA 630, and 0.5% deoxycholate) or samples were eluted from Anti-FLAG M2 magnetic beads (Sigma–Aldrich). For analysis of UPF1 phosphorylation status RIPA buffer was supplemented with 1× PhosSTOP (Roche), 1× Halt^™ Protease and Phosphatase Inhibitor Cocktail (Thermo Fisher Scientific) and 50 μg/μl RNase A (Panreac AppliChem). Protein concentration was determined using the Pierce Detergent Compatible Bradford Assay Reagent (Thermo Fisher Scientific). All antibodies are listed in Supplementary Table S1 and were used at the indicated dilutions in 50 mM Tris (pH 7.2), 150 mM NaCl with 0.2% Tween-20, and 5% skim milk powder. For chemiluminescent detection, membranes were incubated with HRP-conjugated secondary antibodies and visualized using Amersham ECL Prime or Select Western Blotting Detection Reagent (GE Healthcare) in combination with the Fusion FX-6 Edge system (Vilber Lourmat) and Evolution-Capt Edge software (version 18.05) was used for visualization. Quantification of detected protein bands was performed in a semi-automated manner using the Image-Quant TL 1D software (version 8.1) with a rolling-ball background correction. The control condition was set to unity, quantification results are shown as data points and mean.

For analyses with the Odyssey infrared imaging system, dilution series or samples adjusted to 6 µg total protein, prepared in buffers containing additional 0.5 mM NaF and 10 mM β-glycerophosphate disodium salt, were used for electrophoresis and blotted on low fluorescent polyvinylidene fluoride (PVDF) membrane (Thermo Fisher Scientific). After incubation with IRDye-conjugated secondary antibodies, the blots were visualized with the LI-COR Odyssey CLx system using both 700 and 800 nm channel, 169 µm resolution, and intensities set to auto. Quantification was performed using LI-COR Image Studio Version 6.0.0.28 with lane-based background correction. The control condition was set to unity, quantification results are shown as data points and mean (bar).

Stable cell lines and plasmids

For stable integration of FLAG-SMG8/FLAG-SMG9 constructs, the transposon-based PiggyBac (PB) system was used: 2.8 × 10⁵ cells were seeded in six-well plates and after 24 h 1500 ng PB construct and 500 ng PB transposase expressing plasmid were transfected using the calcium phosphate-based method with BES-buffered saline (BBS). Forty-eight hours after transfection, cells were transferred into 10 cm dishes and selected with 100 μg/ml hygromycin (InvivoGen). Colonies were pooled after 10–15 days. Protein expression was induced with 1 μg/ml Cumate. All plasmids used in this study are listed in Supplementary Table S1.

Reverse transcription, end-point RT-PCR

1–4 μg total RNA was used for reverse transcription in a 20 μl of reaction volume with 10 μM VNN-(dT)20 primer using the GoScript Reverse Transcriptase (Promega) following the manufacturer’s instructions. For end-point PCRs, 2% of complementary DNA (cDNA, template), 0.2 μM final concentration of sense and antisense primer (see Supplementary Table S1 for sequences) and MyTaq Red Mix (Bioline) was used. After 30 PCR cycles, the PCR products were resolved by electrophoresis on ethidium bromide-stained, 1% agarose TBE gels, and detected by trans-UV illumination using the Gel Doc XR+ (Bio-Rad) and Image Lab software (version 5.1).

Quantitative RT-PCR, probe-based multiplex RT-PCR

Quantitative RT-PCR was performed with the GoTaq qPCR Master Mix (Promega) using 2% of cDNA in 10 μl reactions, 0.2 μM final concentration of sense and antisense primer (see Supplementary Table S1 for sequences), and the CFX96 Touch Real-Time PCR Detection System (Bio-Rad) with Bio-Rad CFX Manager software (version 3.0). The reactions for each biological replicate were performed in triplicates and the C_t (threshold cycle) value was measured and average C_t values were calculated. For alternative splicing events, values for canonical isoforms were subtracted from values for NMD-sensitive isoforms to calculate the ΔC_t. The mean log2 fold changes were calculated from three biologically independent experiments. Log2 fold change results are shown as data points and mean.

Probe-based multiplex quantitative RT-PCRs were performed using the PrimeTime Gene Expression Master Mix (IDT) and the PrimeTime qPCR Assays containing primers and probes [IDT; SRSF2 = custom (see Supplementary Table S1), ZFAS1 = Hs.PT.58.3673229, GAS5 = Hs.PT.58.24767969, B2M = Hs.PT.58v.18759587, TBP = Hs.PT.58v.39858774] following the manufacturer’s instructions. 2% of cDNA was used as a template in 10 μl reactions and samples were measured using CFX96 Touch Real-Time PCR Detection System (Bio-Rad). The reactions for each biological replicate were performed in triplicates, and the C_t (threshold cycle) value was measured and average C_t values were calculated. The C_t values of the housekeeping gene B2M or TBP (FAM-labeled) were subtracted from the target (ZFAS1, Cy5-labeled, or GAS5, SUN-labeled) values to calculate the ΔC_t. Three biologically independent experiments were used to calculate the mean log2 fold changes. The log2 fold changes are visualized as single data points and mean. All primers used in this study are listed in Supplementary Table S1.

siRNA-mediated knock-downs

2.5–3.0 × 10⁵ cells were seeded in six-well dish and reverse transfected using Lipofectamine RNAiMAX (Invitrogen) and 60 pmol small interfering RNA (siRNA) following the manufacturer’s instructions. Forty-eight hours after transfection cells were harvested in 1 ml in-house prepared TRI reagent [38] for RNA extraction or RIPA buffer (50 mM Tris–HCl pH 8.0, 0.1% SDS, 150 mM NaCl, 1% IGEPAL, and 0.5% deoxycholate) for protein extraction. All siRNAs used in this study are listed in Supplementary Table S1.

High-throughput-sequencing

The RNA was extracted and purified using the Direct-zol RNA MiniPrep kit including the recommended DNase I treatment (Zymo Research; Cat# R2052) according to manufacturer’s instructions. Libraries were prepared from 500 ng total RNA with the Illumina^® Stranded mRNA Preparation kit. ERCC RNA Spike-In Mix (Thermo Fisher) was added to the samples before library preparation. After poly(A) selection [using Oligo(dT) magnetic beads], mRNA was purified, fragmented and reverse transcribed with random hexamer primers. Second strand synthesis with dUTPs was followed by A-tailing, adapter ligation and library amplification (12 cycles) to create the final cDNA libraries. After library validation and quantification (Agilent Tape Station), equimolar amounts of library were pooled. The pool was quantified by using the Peqlab KAPA Library Quantification Kit and the Applied Biosystems 7900HT Sequence Detection System. The pool was sequenced on an Illumina NovaSeq6000 sequencing instrument with a PE100 protocol aiming for 50 million clusters per sample.

Following RNA-Seq datasets were obtained and analyzed: SMG5, SMG6, SMG7 knock-out/knock-down in HEK293 cells (BioStudies [39, 40] accession E-MTAB-9330) [7], and UPF1 degron (AID and dTAG/FKBP) in HEK293 or HCT116 cells (BioStudies accession E-MTAB-13788, E-MTAB-13829) [41].

Computational analyses of RNA-Seq data

For standard RNA-Seq analyses, reads were aligned against the human genome (GRCh38, GENCODE release 42 transcript annotations [42] supplemented with SIRVomeERCCome annotations from Lexogen; obtained from https://www.lexogen.com/sirvs/download/) using the STAR read aligner (version 2.7.10b) [43]. Transcript abundance estimates were computed with Salmon (version 1.9.0) [44] in mapping-based mode using a decoy-aware transcriptome (GENCODE release 42) with the options –numGibbsSamples 30 –useVBOpt –gcBias –seqBias. After the import of transcript abundances in R (version 4.3.0; R Foundation for Statistical Computing. https://www.R-project.org) using tximport (version 1.28.0) [45], differential gene expression (DGE) analysis was performed with the DESeq2 R package (version 1.40.1) [46]. Genes with <10 counts in half the analyzed samples were pre-filtered and discarded. The DESeq2 log2FoldChange estimates were shrunk using the apeglm R package (version 1.22.1) [47]. Differential transcript expression (DTE) analysis was performed using the edgeR R package (version 3.42.4) based on 30 inferential replicate datasets drawn by Salmon using Gibbs sampling, which were used to estimate read-to-transcript ambiguity. After count scaling taking the overdispersion into account, transcripts were pre-filtered using the filterByExpr() function and counts fitted with the quasi-likelihood negative binomial generalized log-linear model. General significance cut-offs, as long as not indicated otherwise, were log2FoldChange > 1 & p.adjust < 0.0001 for DESeq2 DGE and log2FC > 1 & FDR < 0.0001 for edgeR DTE. Gene ontology functional enrichments analysis of gene lists, ordered by adjusted P-value, was performed using an ordered query by g:profiler via the R package gprofiler2 (version 0.2.2) [48], using gene ontology biological process (GO:BP) as data source, a list of all expressed/detected genes as custom background, domain scope set to “custom_annotated” and with “fdr” multiple testing correction method applying significance threshold of 0.05.

Co-immunoprecipitation

Stable cell lines expressing FLAG-tagged SMG8, FLAG-tagged SMG9 or endogenously tagged UPF1 were seeded in 10 cm dishes (2.5–3.0 × 10⁶ cells) and SMG8/SMG9 expression was induced via doxycycline. 2–3 days after seeding cells were harvested in 200 μl buffer E phos [20 mM HEPES-KOH (pH 7.9), 100 mM KCl, 10% glycerol, 1 mM dithiothreitol (DTT), Protease Inhibitor, 1× PhosSTOP (Roche), 1× Halt^™ Protease and Phosphatase Inhibitor Cocktail (Thermo Scientific), and 50 μg/μl RNase (Panreac AppliChem)]. Cells were lysed using Bandelin Sonopuls mini20 with 15 pulses (2.5 mm tip, 1 s pulse, 50% amplitude). Samples were adjusted to the same concentration and incubated for 2 h overhead shaking with Anti-FLAG M2 Magnetic Beads (Sigma–Aldrich). Beads were washed three times for 5 min with mild wash buffer [20 mM HEPES-KOH (pH 7.9), 137 mM NaCl, 2 mM MgCl₂, 0.2% Triton X-100, and 0.1% NP-40). Co-immunoprecipitated proteins were eluted with SDS-sample buffer, separated by SDS–PAGE, and analyzed by western blotting.

Label-free quantitative mass spectrometry

Cells expressing endogenously FLAG-tagged UPF1 were seeded in 10 cm dishes (2.5–3.0 × 10⁶ cells). After 24 h, cells were treated with SMG1i [49] for 24 h and harvested in 200 μl buffer E phos [20 mM HEPES-KOH (pH 7.9), 100 mM KCl, 10% glycerol, 1 mM DTT, Protease Inhibitor, 1× PhosSTOP (Roche), 1× Halt^™ Protease and Phosphatase Inhibitor Cocktail (Thermo Scientific), and 50 μg/μl RNase (Panreac AppliChem)] and immunoprecipitation was performed as described above using mild wash buffer. Proteins were eluted using 44 µl FLAG-peptides (200 μg/ml; Merck/Sigma–Aldrich) in 1× TBS. 44 μl of 10% SDS in 1× PBS was added and samples were incubated at 95 °C for 5 min. Samples were reduced with DTT at 55 °C for 30 min and alkylated with CAA at RT for 30 min (final concentrations 5 and 55 mM, respectively).

Tryptic protein digestion was achieved by following a modified version of the single pot solid phase‐enhanced sample preparation (SP3) [50]. In brief, paramagnetic Sera‐Mag speed beads (Thermo Fisher Scientific) were added to the reduced and alkylated protein samples and then mixed 1:1 with 100% acetonitrile (ACN). Protein‐beads complexes form during the 8‐min incubation step, followed by capture using an in‐house build magnetic rack. After two washing steps with 70% EtOH, the samples were washed once with 100% ACN. Then they were air‐dried, resuspended in 5 μl of 50 mM triethylamonium bicarbonate supplemented with trypsin in an enzyme:substrate ratio of 1:50 and incubated for 16 h at 37 °C. Afterwards, samples were acidified to 5% FA and cleaned-up using SDB-RPS StageTips. Samples were loaded onto the tips, washed with 0.1% FA followed by washing with 80% AcN + 0.1% FA. Finally, samples were eluted with 40% NH₃, dried down and resuspended in 4% AcN + 0.1% FA, ready for mass spectrometric analysis.

Data acquisition

Samples were analyzed by the CECAD Proteomics Facility on an Orbitrap Exploris 480 (Thermo Scientific, granted by the German Research Foundation under INST 216/1163-1 FUGG) mass spectrometer equipped with a FAIMSpro differential ion mobility device that was coupled to an Vanquish neo in trap-and-elute setup (Thermo Scientific). Samples were loaded onto a precolumn (Acclaim 5µm PepMap 300 µ Cartridge) with a flow of 60 µl/min before reverse-flushed onto an in-house packed analytical column (30 cm length, 75 µm inner diameter, filled with 2.7 µm Poroshell EC120 C18, Agilent). Peptides were chromatographically separated with an initial flow rate of 400 nl/min and the following gradient: initial 2% B (0.1% formic acid in 80% acetonitrile), up to 6% in 3 min. Then, flow was reduced to 300 nl/min and B increased to 20% B in 26 min, up to 35% B within 15 min and up to 98% solvent B within 1.0 min while again increasing the flow to 400 nl/min, followed by column wash with 95% solvent B and re-equilibration to initial condition. The FAIMS pro was operated at −40 V compensation voltage and electrode temperatures of 99.5 °C for the inner and 85 °C for the outer electrode. The mass spectrometer was operated in data-dependent acquisition top 24 mode with MS1 scans acquired from 350 m/z to 1400 m/z at 60k resolution and an AGC target of 300%. MS2 scans were acquired at 15k resolution with a maximum injection time of 22 ms and an AGC target of 300% in a 1.4 Th window and a fixed first mass of 110 m/z. All MS1 scans were stored as profile, all MS2 scans as centroid.

Sample processing in MaxQuant

All mass spectrometric raw data were processed with MaxQuant (version 2.4) [51] using default parameters against the UniProt Human canonical reference proteome database (UP5640) with the match-between-runs option enabled between replicates. Label-free quantification was performed separately for replicate group to better cope with strong differences in protein abundances in IP situations. Follow-up analysis was done in Perseus 1.6.15 [52]. Protein groups were filtered for potential contaminants and insecure identifications. Remaining IDs were filtered for data completeness in at least one group and missing values imputed by sigma downshift (0.3 σ width, 1.8 σ downshift). Afterwards, FDR-controlled two-sided t-tests were performed. Finally, majority protein IDs were used for protein annotations and further analysis.

Data presentation

Schematic representations and figures were prepared/assembled using CorelDraw 2017 or 2025. Quantifications and calculations for other experiments were performed - if not indicated otherwise - with Microsoft Excel (version 1808 or 2311) or R (version 4.3.0) and all plots were generated using IGV (version 2.14.1) [53], GraphPad Prism 5, ggplot2 (version 3.4.2) [54], ggsashimi (version 1.1.5) [55], nVennR (version 0.2.3) [56], or ComplexHeatmap (version 2.18.0) [57]. If not indicated otherwise, the box of boxplots extends to the 25th and 75th percentile with the median in bold line, outliers are not shown.

Quantification and statistical analysis

Most performed statistical tests are already implemented in the used bioinformatic tools. For DGE analysis, P-values were calculated by DESeq2 using a two-sided Wald test and corrected for multiple testing using the Benjamini–Hochberg method. For DTE analysis, P-values were calculated by edgeR using transcriptwise negative binomial generalized linear models with quasi-likelihood tests and corrected for multiple testing using FDR approaches. Linear regression of scatter plots was performed using the stat_poly_eq function of the ggpmisc R package, displaying the adjusted coefficient of determination. No blinding approach was applied to the analyses.

Results

The KID of SMG8 is dispensable for NMD

A pivotal step during the initiation of NMD is the phosphorylation of PTC-proximal UPF1 molecules by the SMG1:8:9 complex, marking aberrant transcripts for degradation (Fig. 1A). This phosphorylation enables recruitment of the SMG5:SMG7 complex and subsequent endonucleolytic cleavage via the composite SMG5-SMG6 PIN domain. Although SMG1 kinase activity is central to this process, the regulatory mechanisms controlling its activation and substrate specificity remain only partially understood. Structural and biochemical studies have proposed that SMG8 serves as a negative regulator of SMG1, acting through its C-terminal KID to suppress SMG1 activity in vitro [16, 30–33, 35]. However, whether this regulatory function is essential in vivo remains unexplored. Based on these previous findings, we hypothesized that the deletion of the KID would lead to increased UPF1 phosphorylation and altered NMD activity in cells. To test this hypothesis, we generated HCT116 cells in which a cassette containing a Myc tag, a Puromycin resistance marker, and poly(A) signal was inserted into the second exon of the endogenous SMG8 locus using the CRISPaint method [36] (Supplementary Fig. S1A and Supplementary Table S1). This modification effectively truncated the protein upstream of the KID, as validated by western blot analysis (Fig. 1B). Co-immunoprecipitation analyses demonstrated that the truncated SMG8 retained its ability to interact with SMG9, indicating that removal of the KID does not disrupt SMG8–SMG9 complex formation in vivo and confirming the absence of residual full-length SMG8 protein in KID-deleted cells (Fig. 1C). However, the UPF1 phosphorylation level as detected by a phospho-UPF1 specific antibody (serine 1116; short loop isoform, UniProt ID: Q92900-2) was only marginally altered in the KID-deleted cells (Fig. 1D). To assess the consequences of KID removal, we sequenced poly(A)+ enriched mRNA isolated from the engineered cells. The RNA-seq analysis confirmed the deletion of the KID at the mRNA level, with no detectable full-length SMG8 mRNA (Supplementary Fig. S1B). DGE analysis revealed overall modest effects with more significantly downregulated than upregulated protein-coding genes (Fig. 1E). At the transcript level, only a very small subset of NMD-annotated transcripts was upregulated, indicating that the KID deletion did not significantly impact NMD efficiency (Fig. 1F). Similarly, only few long noncoding RNA (lncRNA)-annotated transcripts were upregulated, many of which harbour putative open reading frames and have been reported to be regulated by NMD [41] (Fig. 1F). Moreover, canonical NMD-sensitive genes such as GADD45B, ZFAS1, and GABARAPL1 [58–60] remained unaffected (Fig. 1E). Gene ontology analysis did not uncover any significant pathways associated with the observed differential expression patterns (Supplementary Fig. S1C). Overall, these findings indicate that the SMG8 KID is not required for controlling UPF1 phosphorylation or for sustaining NMD activity in human cells. This suggests that the inhibitory role of SMG8 observed in vitro may not fully capture its function in a cellular context, emphasizing the need for in vivo studies to better understand the regulatory mechanisms within the NMD pathway.

Figure 1.

NMD activity is unaffected by the deletion of the SMG8 KID. (A) Schematic representation of the central steps of NMD. The SMG1 kinase is regulated by SMG8 and SMG9. The SMG1:SMG8:SMG9 complex is recruited to mRNA-bound UPF1 and phosphorylates the N- and C-terminal tails of UPF1 allowing the heterodimer SMG5:SMG7 and the endonuclease SMG6 to bind. SMG5:SMG7 activate SMG6, resulting in the endonucleolytic cleavage of the mRNA in the vicinity of the premature termination codon (PTC) via SMG6. (B) Western blot analysis of cells with deleted KID of SMG8 (delKID; retaining amino acids M1-P597). Anti-Myc (AK-106) and anti-SMG8 (AK-159) antibodies were used (n = 1 biologically independent sample; see Supplementary Table S1 for antibody details). The region of SMG8 detected by the SMG8 antibody is schematically depicted. (C) Western blot analysis of co-immunoprecipitation of FLAG-tagged GST (control) or SMG9 in WT or SMG8 delKID cells. TCE-staining served as loading control. Anti-SMG8 (AK-159) and anti-FLAG (AK-115) antibodies were used (n = 1 biologically independent sample). (D) Analysis of UPF1 serine 1116 (corresponding to the UPF1 short loop isoform; UniProt ID: Q92900-2) phosphorylation status in WT and SMG8 delKID cells. TCE-staining serves as a control. Quantification of total UPF1 (anti-UPF1; AK-156) or phosphorylated UPF1 (anti-pUPF1; serine 1116; AK-146) is normalized to one representative TCE-staining and is shown as data points and mean (n = 3 biologically independent samples). (E) Ridge plot showing the DGE in SMG8 delKID versus control RNA-seq data. Significantly regulated (p.adjust < 0.0001) genes with GENCODE-annotated gene biotypes protein-coding, lncRNA, or other are shown. The log2 fold change in gene expression is plotted against the density height scaled by counts. Individual known NMD-targeted genes are highlighted. (F) Ridge plot showing the DTE in SMG8 delKID versus control RNA-seq data. Significantly regulated (FDR < 0.0001) transcripts with GENCODE-annotated transcript biotypes protein-coding, lncRNA, NMD, or other are shown. The log2 fold change in transcript expression is plotted against the density height scaled by counts.

SMG8 or SMG9 function as modulators rather than essential NMD effectors

Since deletion of the SMG8 KID did not result in significant changes in UPF1 phosphorylation or NMD efficiency, we next sought to explore the broader roles of SMG8 and SMG9 in regulating SMG1 activity. Both proteins have been implicated as negative regulators of SMG1, and are thought to modulate its kinase function within the SMG1:8:9 complex. We therefore reasoned that complete inactivation of either SMG8 or SMG9 would be required to reveal their regulatory impact, potentially resulting in elevated UPF1 phosphorylation and measurable changes in NMD activity (Fig. 2A). To test this, we generated knockout (KO) lines for SMG8 and SMG9 in HCT116 cells using the CRISPaint system [36]. Successful gene disruption was confirmed by Sanger sequencing and western blotting (Fig. 2B). Co-immunoprecipitation analyses confirmed that the edited SMG9–GFP fusion protein (retaining SMG9 residues 1–135) failed to interact with SMG8, rendering this truncated protein nonfunctional (Supplementary Fig. S2A).

Figure 2.

Weak effects on NMD activity in SMG8- and SMG9-deficient cells. (A) Schematic overview of the SMG1:SMG8:SMG9 complex without SMG8 or SMG9. The lack of SMG8 or SMG9 might influence UPF1 phosphorylation resulting in altered NMD activity. (B) Western blot analysis of SMG8 or SMG9 KO cells using anti-SMG8 (AK-169) and anti-SMG9 (AK-170) antibodies. Tubulin (AK-084) serves as control (n = 1 biologically independent sample; see Supplementary Table S1 for antibody details). Asterisk indicates nonspecific bands. Domain structure of full-length SMG8 and SMG9 protein and truncated proteins of SMG8 and SMG9 KO cells are shown. The region detected by the antibodies is schematically depicted. (C) End-point RT-PCR detection of SRSF2 transcript isoforms (top) and quantitative probe-based RT-PCR (bottom) of SRSF2 (NMD isoform) and ZFAS1 in WT, SMG8 KO or SMG9 KO cells with or without indicated knock-downs (KD). The detected SRSF2 isoforms are indicated on the right (NMD = NMD-inducing isoform; canon. = canonical isoform). For the probe-based RT-qPCR, the ratio of SRSF2 to the B2M reference, and ZFAS1 to the TBP reference was calculated; data points and means from the qPCRs are plotted as log2 fold change (log2FC) (n = 3 biologically independent samples). Western blot analysis of SMG6 and SMG7 KD efficiency is shown with the anti-SMG6 (AK-135) and anti-SMG7 antibody (AK-136). Tubulin (AK-084) serves as a control (n = 1 biologically independent sample; see Supplementary Table S1 for antibody details). (D) Comparison of SMG8 KO, SMG9 KO, and SMG8 delKID RNA-Seq data with SMG7 KO + SMG6 KD (clone 34) [7] or three UPF1 degron conditions [41] regarding the log2FC distribution and number of significantly up- or downregulated GENCODE-annotated genes (p.adjust < 0.0001; left) or significantly up- or downregulated GENCODE NMD-annotated transcripts (FDR < 0.0001; right). (E) Read coverage of SRSF2 from SMG8 KO, SMG9 KO, and SMG8 delKID and published SMG7 KO + SMG6 KD (clone 34) [7]. RNA-Seq data is shown as Integrative Genomics Viewer (IGV) snapshots (n = 3 biologically independent samples). The canonical and NMD-sensitive isoforms are schematically indicated below. (F and G) Overlaps between differentially regulated genes (F) or transcripts (G) of SMG8 and SMG9 KO cells. Scatter plots show the change in gene (F) or transcript(G) expression of SMG8 KO cells against the change in SMG9 KO cells (DGE: p.adjust < 0.0001; DTE: FDR < 0.0001). Linear regression with P-value (P) and adjusted coefficient of determination is shown (n = 3 biologically independent samples).

To test how the depletion of SMG8 and SMG9 influences NMD activity, we analyzed the well-known exon inclusion event for the NMD target SRSF2 (Supplementary Fig. S2B) and the abundance of the two NMD-sensitive lncRNAs GAS5 and ZFAS1 [58, 61]. All three NMD targets exhibited low accumulation compared to co-depletion of the NMD factors SMG6 and SMG7 that severely abolish NMD activity (Fig. 2C and Supplementary Fig. S2C). Hence, SMG8 and SMG9 KO cells exhibit a weak to moderate NMD inhibition.

RNA-seq of the SMG8-depleted HCT116 cells verified the complete loss of the SMG8 mRNA (Supplementary Fig. S2D). The SMG9 KO cells expressed almost normal levels of the expected shortened transcript as well as low levels of alternatively spliced SMG9 mRNA (Supplementary Fig. S2E). Almost all of these exon-skipping transcripts lead to frame shifts resulting in truncated and presumable nonfunctional transcripts (Supplementary Fig. S2F). DGE and DTE analysis of NMD-annotated transcripts revealed weak transcriptome-wide effects in SMG8 or SMG9 KO cells compared to conditions that induce strong NMD inhibition [SMG7 KO + SMG6 knock-down (KD)] [7] or degron-mediated UPF1 depletion (AID/FKBP-UPF1) [41]) (Fig. 2D and E). Comparison of significantly altered genes and transcripts in SMG8- and SMG9-depleted cells revealed 1043 shared DGE and 575 shared DTE events with good correlation between conditions (Fig. 2F-G). In contrast, the shared DGE events of SMG8 KO and SMG8 delKID cells displayed poor correlation, reinforcing the conclusion that the KID is dispensable for NMD function (Supplementary Fig. S2G). Taken together, loss of SMG8 or SMG9 resulted in mild NMD impairment suggesting a modulatory rather than an essential role in regulating SMG1 activity.

SMG8 or SMG9 knockout causes modest increases in UPF1 phosphorylation

Based on the mild NMD effects resulting from SMG8 or SMG9 depletion, we asked whether this effect reflects changes in UPF1 phosphorylation, employing a Ser1116 phospho-specific UPF1 antibody as a readout (Fig. 3A). We quantified total and Ser1116-phosphorylated UPF1 using the Odyssey infrared imaging system, which provides linear, quantitative protein detection without enzymatic amplification steps. Dual-wavelength imaging permitted simultaneous measurement of UPF1 and phospho-UPF1 (Ser1116) on the same blot, and incorporation of a dilution series ensured linearity of signal detection (Fig. 3B and C). We detected increased phospho-UPF1 to total UPF1 ratios in two independent SMG8 KO and two independent SMG9 KO clones, although the effect was overall more pronounced in the absence of SMG9, especially clone A (Fig. 3D). As a control for altered UPF1 phosphorylation, treatment with the SMG1 inhibitor (SMG1i; compound 11j) [49] strongly reduced phospho-UPF1 levels (Fig. 3B, lane 11). SMG1i functions as an ATP-competitive inhibitor that binds to the active site of SMG1 and stabilizes its autoinhibitory conformation [33]. Together, these results suggest that loss of SMG8 or SMG9 leads to varying degrees of dysregulation of UPF1 phosphorylation, albeit with only limited impact on NMD (Fig. 2).

Figure 3.

Altered UPF1 phosphorylation and functional consequence in SMG8 and SMG9 KO cells. (A) Schematic representation of the UPF1 domain structure (short loop isoform; UniProt ID: Q92900-2). Positions of [S/T]Q motifs are indicated with black lines and black font. L[S/T]Q motifs are shown with black lines and yellow font. The epitope recognized by the pUPF1 antibody (serine 1116; AK-146) is indicated. (B) Two-channel fluorescent western blot analysis of UPF1 phosphorylation status in WT (dilution series), SMG8 or SMG9 KO cells detecting simultaneously total UPF1 (anti-UPF1; AK-156; secondary IRDye 680RD-conjugated antibody) and phosphorylated UPF1 (anti-pUPF1; serine 1116; AK-146; secondary IRDye 800CW-conjugated antibody). TCE-staining served as loading control. SMG1i treatment (WT S1i) for 24 h was used as negative control. Signal of each channel was quantified and normalized to 6 µg loaded WT control (n = 4 biologically independent sample; see Supplementary Table S1 for antibody details). (C) Plot showing the normalized signal versus normalized load for the dilution series of each replicate and channel, the adjusted coefficient of determination is indicated. (D) Comparison of log2FC of pUPF1/UPF1 ratios from different conditions. (E) Schematic representation of SMG1 inactivation via siRNA-mediated knock-down (KD) or treatment with the SMG1 inhibitor SMG1i and subsequent expected changes in UPF1 phosphorylation and NMD activity. (F–H) End-point RT-PCR detection of SRSF2 transcript in WT, SMG8 KO or SMG9 KO cells with Luc or SMG1 KD (F); WT cells (G) or different cell lines (H) treated with various concentration of SMG1i for 24 h. The detected SRSF2 isoforms are indicated on the right (NMD = NMD-inducing isoform; canon. = canonical isoform). Quantitative probe-based RT-PCR of SRSF2 was performed for (H). For probe-based RT-qPCR, the ratio of SRSF2 to the B2M reference was calculated; data points and means from the qPCRs are plotted as log2 fold change (log2FC) (n = 3 biologically independent samples).

Increased sensitivity to SMG1 inhibition in SMG8- or SMG9-deficient cells

The previously reported regulatory role of SMG8 and SMG9 suggests that loss of either factor results in increased SMG1 activity (Fig. 3E). In line with this notion, SMG8- and SMG9-deficient cells displayed elevated UPF1 phosphorylation (Fig. 3D). To attenuate the apparent hyperactivation of SMG1, we performed siRNA-mediated SMG1 knockdown, anticipating normalization of UPF1 phosphorylation and recovery of NMD activity. Instead, SMG1 depletion resulted in marked accumulation of SRSF2 and moderate increases in ZFAS1 and GAS5 exclusively in SMG8- and SMG9-deficient cells, indicating that these cells are uniquely sensitive to reduced SMG1 levels (Fig. 3F and Supplementary Fig. S3A). To determine if this effect was caused by the absence of the SMG1 protein itself or by the lack of its kinase function, we next employed the SMG1i [49]. In wild-type cells, treatment with 0.5 µM SMG1i caused pronounced NMD inhibition accompanied by marked UPF1 hypophosphorylation (Fig. 3G and Supplementary Fig. S3B). SMG8 and SMG9 KO cells exhibited severe NMD inhibition already at 0.1 µM SMG1i, revealing a heightened sensitivity to SMG1 inhibition in the absence of either regulatory subunit (Fig. 3H and Supplementary Fig. S3C). To directly assess UPF1 phosphorylation, we used cells in which UPF1 was endogenously FLAG-tagged, enabling enrichment of UPF1 by FLAG-immunoprecipitation (Supplementary Fig. S3D). Phosphorylation was then evaluated using a UPF1-specific phospho-antibody and a broad-spectrum phosphor-Ser/Thr antibody. Using this strategy, we found that UPF1 phosphorylation levels were comparable between WT and KO cells upon treatment with 0.1 µM SMG1i, indicating that the severe NMD impairment observed cannot be explained solely by alterations in global UPF1 phosphorylation status (Supplementary Fig. S3E; compare lanes 13, 16, and 19). Increasing the inhibitor concentration to 1 µM further reduced phosphorylation of UPF1 serine 1116 in all cell lines (Supplementary Fig. S3E; compare lanes 14, 17, and 20). Together, these findings demonstrate that NMD execution in SMG8- and SMG9-deficient cells is acutely vulnerable to partial SMG1 inhibition, highlighting an essential buffering role for SMG8 and SMG9 in maintaining robust SMG1 function.

SMG1i treatment results in concentration-dependent NMD inhibition

Given the strong accumulation of NMD-sensitive transcripts of SRSF2, ZFAS1, and GAS5 upon SMG1 inactivation via SMG1i, we next sought to investigate the transcriptome-wide effects caused by this treatment. Principal component analysis (PCA) of RNA-seq data revealed a clear, dose-dependent trend along principal component 1 (PC1), reflecting a progressive transcriptome-wide response to increasing concentrations of SMG1i (Fig. 4A). Further investigation of the top 100 genes contributing to the PCA revealed that PC1 captured largely NMD-relevant effects (Supplementary Fig. S4A). This indicates that the SMG1i treatment influences transcriptome-wide changes more profoundly than the SMG8/SMG9 KO or SMG8 delKID alone. The delKID cells showed a similar distribution compared to WT cells, however, were shifted along principal component 2. Importantly, genes previously found to be dysregulated in SMG8 delKID cells (Fig. 1E) were not normalized by SMG1i treatment (Supplementary Fig. S4B), suggesting that these effects are likely independent of SMG1 activity or NMD.

Figure 4.

NMD is strongly inhibited transcriptome-wide after SMG1 inactivation. (A) Principal component analysis of gene-level counts from RNA-seq data of WT, SMG8 KO, SMG9 KO, and SMG8 delKID cells treated with different SMG1i concentrations for 24 h. Lines were added to visualize the samples from the same cell line, color shades indicate SMG1i concentrations. (B) RNA-seq data of WT, SMG8 KO, SMG9 KO, and SMG8 delKID cells treated with different concentrations of SMG1i for 24 h were compared with SMG7 KO + SMG6 knock-down (KD; clone 34) [7] or three UPF1 degron conditions [41] regarding the log2 fold change of significantly regulated NMD-annotated transcripts (FDR < 0.0001; left) and number of significantly upregulated NMD-annotated transcripts (FDR < 0.0001 & log2FC > 1; right), based on the GENCODE reference annotation. (C) Heatmap of NMD-annotated transcript expression changes in the indicated conditions, based on the NMD-regulated human transcriptome (NMDRHT) annotation [41], sorted by effect in UPF1 AID degron condition. (D) Schematic overview of potential consequences of SMG1 complex and/or activity modulation on the UPF1 interactome. (E) Heatmap of mean log2 intensities of the indicated proteins measured by mass spectrometry after co-immunoprecipitation of endogenously FLAG-tagged UPF1 (n = 4 biologically independent samples). Samples were derived from untagged (control), FLAG-UPF1 expressing WT or SMG8/SMG9 KO cells, treated with the indicated concentrations of SMG1i for 24 h. Lysates contained 50 μg/μl RNase A. (F) Scatter plot of mass spectrometry measured mean log2 intensities highlighting NMD factors and comparing different conditions to untreated WT cells. Missing values were substituted with log2(intensity) = 10 for visualization.

Analysis of significantly regulated NMD-annotated transcripts confirmed the increased sensitivity of SMG8 and SMG9 KO cells to SMG1 inhibition. Even at low concentrations of SMG1i, these cells exhibited substantially more and stronger upregulated NMD-targeted transcripts compared to wild-type cells (Fig. 4B). Furthermore, high concentrations of SMG1i (1 μM) resulted in extensive NMD suppression, similar to the effects observed with SMG6 and SMG7 co-depletion or UPF1 depletion via degron tags. Remarkably, this global inhibition of NMD occurred without major changes in the expression of core NMD factors (Supplementary Fig. S4C).

For further in-depth investigation we repeated the DTE analysis using the consolidated NMD-regulated human transcriptome (NMDRHT) annotation [41]. Comparable to the GENCODE-based analyses, the high SMG1i concentration (1 μM) treatment largely phenocopied the effects of UPF1 depletion (Fig. 4C), irrespective of NMD target classification based on activating properties (NMD reason), temporal clustering (DTE cluster) or NMD relevance (Supplementary Fig. S4D–F). Accordingly, SMG8 and SMG9 KO cells treated with low SMG1i concentration (0.1 μM) showed intermediate effects across all classes of NMD-targeted transcripts (Fig. 4C and Supplementary Fig. S4D–F). These results indicate that SMG8 or SMG9 deficiency lowers the threshold for further NMD perturbations, resulting in global dysregulation of the NMD targeted transcriptome.

Catalytically inactive SMG1:SMG8:SMG9 complexes affect composition of the NMD machinery

Given the transcriptome-wide effects on NMD, we hypothesized that alterations in SMG1:SMG8:SMG9 complex composition and/or activity influence the composition of the UPF1 interactome (Fig. 4D). To investigate this, we performed immunoprecipitation of endogenously FLAG-tagged UPF1 followed by label-free mass spectrometry. Compared to untagged control, FLAG-UPF1 pull-downs from untreated WT cells were enriched for most core NMD factors, several EJC components, most notably CASC3, and other known interactors such as STAU2, MOV10, and SZRD1 (Fig. 4E and Supplementary Fig. S5A).

In untreated SMG8 or SMG9 KO cells, the composition of NMD-relevant UPF1-containing complexes was largely unchanged, although SMG1 levels were markedly increased (Fig. 4E, and Supplementary Fig. S5A and B). Treatment with low concentrations of SMG1i (0.1 µM) further enhanced the association of SMG1 with UPF1 in the SMG8 and SMG9 KO cells, which was also observed when treating wild type cells with high SMG1i (1 µM) concentrations (Fig. 4F). Depending on the genotype (WT or KO), also SMG8 and/or SMG9 were found to be more associated with UPF1 in these conditions. Of note, the majority of detected SMG9 peptides were derived from the N-terminal region of SMG9, thereby accounting for the residual intensity measured in SMG9 KO cells (Supplementary Fig. S5C).

Additionally, treatment with high concentrations of SMG1i led to a decrease in SMG5 and SMG7 association with UPF1, in line with their phosphorylation-dependent recruitment (Fig. 4E and F, and Supplementary Fig. S5A and B). Interestingly, SMG6 peptides were detected in UPF1 co-immunoprecipitates exclusively under severely NMD-compromised conditions (high SMG1i in WT or low SMG1i in SMG8/9 KO), reinforcing the capacity of SMG6 to bind UPF1 independently of its phosphorylation status [21, 22, 62]. Combined with the RNA-seq analyses, the proteomic results suggest that SMG1 complexes compromised by catalytic inhibition or loss of regulatory factors (SMG8 or SMG9) leads to the unproductive assembly of the NMD machinery, which cannot execute RNA degradation.

Loss of SMG8 or SMG9 reveals reproducible effect in NMD across multiple cell lines

To assess whether the effects of SMG8 or SMG9 KO on UPF1 phosphorylation and sensitivity to partial SMG1 inhibition are conserved across cellular contexts, we extended our analysis to U2OS and HEK293 cells (Fig. 5A and B). Consistent with our observations in HCT116 cells, Odyssey infrared imaging revealed increased Ser1116 UPF1 phosphorylation in SMG8 and SMG9 KO cells across all cell lines, with the notable exception of SMG8 KO U2OS cells (Fig. 5C and Supplementary Fig. S6A–F). Overall, the increase in UPF1 phosphorylation upon SMG8/SMG9 KO was not as pronounced in HEK293 and U2OS as in HCT116 cells.

Figure 5.

Reproducible effects on NMD activity in U2OS and HEK293 SMG8 and SMG9 KO cells. (A) Schematic overview of HEK293 and U2OS cell lines used for SMG1 complex and activity modulation in comparison to previous results obtained in HCT116 cells. (B) Western blot analysis of SMG8 or SMG9 KO HEK293 (left) and U2OS (right) cells using anti-SMG8 (AK-169) and anti-SMG9 (AK-170) antibodies. TCE-staining serves as loading control (n = 1 biologically independent sample; see Supplementary Table S1 for antibody details). (C) Comparison of log2FC of pUPF1/UPF1 ratios determined by two-channel fluorescent western blot from different cell lines and conditions, merging the results of individual clones. (D) End-point RT-PCR detection of SRSF2 transcripts in WT, SMG8 KO, or SMG9 KO HEK293 (left) and U2OS (right) cells treated with different concentration of SMG1i for 24 h. The detected SRSF2 isoforms are indicated on the right (NMD = NMD-inducing isoform; canon. = canonical isoform; n = 3 biologically independent samples). (E) Comparison of different RNA-seq data by heatmap of median log2 fold change of transcripts stratified by potential NMD activating reason, based on the NMD-regulated human transcriptome (NMDRHT) annotation [41]. AS-NMD = alternative splicing coupled to NMD; AS-NMD-UTR3 = NMD-inducing splicing events in the 3′ UTR; novel = ORF or transcript not found in reference annotation; uORF = upstream open reading frame. (F) UpSet plot of the overlap of significantly upregulated transcripts (FDR < 0.0001 & log2FC > 1) from severely NMD-inhibited conditions, stratified by NMDRHT-based NMD annotation (TRUE/FALSE). NA = no information about open reading frame.

Functionally, treatment of SMG8 and SMG9 KO cells with low concentration of SMG1i (0.1 µM) resulted in upregulation of the SRSF2 NMD isoform in both HEK293 and U2OS cell line, mirroring the results obtained in HCT116 cells (Fig. 5D and Supplementary Fig. S7A). This suggests a similar sensitization of cells lacking SMG8 or SMG9 to additional perturbations of SMG1 activity, even in cells where UPF1 phosphorylation is only weakly increased or unchanged (Fig. 5C and Supplementary Fig. S6D–F).

To further characterize this effect at the transcriptome-wide level, we performed RNA-seq analysis of U2OS WT, SMG8 KO, and SMG9 KO cells treated with increasing concentrations of SMG1i. Principal component analysis revealed that most of the variance across conditions is explained by an NMD-relevant principal component 1 (PC1), which separated untreated or 0.1 μM-treated WT cells from conditions of strong NMD inhibition, including high SMG1i treatment or low-dose SMG1i treatment in SMG8 or SMG9 KO cells (Supplementary Fig. S7B and C). DTE analysis confirmed the substantial NMD inhibition in U2OS SMG8 or SMG9 KO cells with low SMG1i treatment, consistent with results obtained in HCT116 cells (Fig. 5E, Supplementary Fig. S7D). Comparison of strongly inhibited NMD conditions revealed a substantial overlap of NMD-annotated upregulated transcripts (Fig. 5F). Collectively, these results demonstrate that the functional NMD impairment caused by SMG8 or SMG9 loss is reproducible across distinct human cell lines.

Synergy between SMG1:SMG8:SMG9 complex disruption and additional NMD perturbations

Our previous findings establish that SMG8 and SMG9 modulate SMG1 function and contribute to NMD efficiency, despite not being strictly essential for NMD activity. We next asked whether disruption of the SMG1:SMG8:SMG9 complex sensitizes NMD to further perturbations, thereby revealing functional dependencies that are otherwise buffered under intact conditions. To address this, we employed two complementary approaches. First, we combined a low concentration of SMG1i (0.1 µM) with siRNA-mediated knockdown of SMG6 in wild-type HCT116 cells. This dual perturbation led to substantial accumulation of the NMD target SRSF2, indicating a strong inhibition of NMD (Fig. 6A and Supplementary Fig. S8A). Similarly, SMG6 knockdown in SMG8 or SMG9 KO cells resulted in pronounced NMD inhibition, most clearly observed with SRSF2 (Fig. 6A and Supplementary Fig. S8A). These observations suggest that partial disruption of SMG1 complex regulation sensitizes the pathway to additional interference. To examine whether this effect extends to other partially impaired NMD backgrounds, we treated HEK293 cells lacking the NMD factor SMG7 or EJC-component CASC3 with low-dose SMG1i. Both genes are known to contribute to moderate NMD impairment [7, 63]. In both cases, SMG1 inhibition further reduced NMD activity, resulting in effects similar to what is seen in severe NMD knockdowns (Fig. 6B and Supplementary Fig. S8B). These findings indicate that even modest perturbation of the SMG1:SMG8:SMG9 complex renders the NMD pathway more susceptible to additional disruptions. This increased vulnerability highlights a fragility within the NMD machinery, where partial impairments at one level can unmask otherwise buffered roles of additional factors. This phenomenon could serve as a powerful tool to dissect the contributions of nonessential NMD components.

Figure 6.

Impact of SMG1:SMG8:SMG9 complex impairment on NMD parameters. (A) End-point RT-PCR detection of SRSF2 transcript isoforms (top) and quantitative probe-based RT-PCR (bottom) of SRSF2 and ZFAS1 in WT, SMG8 KO and SMG9 KO cells with Luc knock-down (KD; control) or SMG6 KD. WT cells were treated in addition with SMG1i for 24 h. The detected SRSF2 isoforms are indicated on the right (NMD = NMD-inducing isoform; canon. = canonical isoform). For probe-based RT-qPCR, the ratio of SRSF2 or ZFAS1 to the B2M reference was calculated; data points and means from the qPCRs are plotted as log2 fold change (log2FC) (n = 3 biologically independent samples). Western blot analysis of SMG6 KD efficiency is shown with the anti-SMG6 antibody (AK-135). TCE-staining serves as a control (see Supplementary Table S1 for antibody details). (B) End-point RT-PCR detection of SRSF2 transcript isoforms (top) and quantitative probe-based RT-PCR (bottom) of ZFAS1 in WT, SMG8 KO, SMG7 KO, and CASC3 KO cells. Cells were treated in addition with the indicated concentration of SMG1i for 24 h. The detected SRSF2 isoforms are indicated on the right (NMD = NMD-inducing isoform; canon. = canonical isoform). For probe-based RT-qPCR, the ratio of ZFAS1 to the TBP reference was calculated; data points and means from the qPCRs are plotted as log2 fold change (log2FC) (n = 3 biologically independent samples). (C) Left: The SMG1:SMG8:SMG9 complex phosphorylates UPF1, initiating the first “authentication” step of NMD. Phosphorylated UPF1 recruits SMG5:SMG7, which activates SMG6 to cleave the mRNA and promote complex disassembly. In the absence of SMG8 or SMG9, SMG1 can still phosphorylate UPF1 and allow downstream activation, but the system is partially compromised (not shown). This state renders the pathway more sensitive to additional perturbation. Right: Low-dose SMG1 inhibitor (SMG1i) partially inhibits SMG1i kinase activity. This synergizes with SMG8 or SMG9 depletion, as well as with the loss of SMG7 or CASC3, leading to enhanced NMD inhibition. Proteins outlined with a red solid line represent essential NMD factors, while those outlined with a green dashed line denote redundant factors.

Discussion

The phosphorylation of UPF1 is widely recognized as a hallmark event in NMD, long considered a definitive commitment point in the pathway [7, 64–66]. Yet, the molecular process of UPF1 phosphorylation by SMG1 and its cofactors SMG8 and SMG9 remains incompletely understood. In this study, we sought to clarify the physiological roles of SMG8 and SMG9 in regulating UPF1 phosphorylation and NMD activity. Beyond their mechanistic interest, SMG8 and SMG9 are clinically relevant: mutations in either gene are linked to congenital developmental disorders marked by organ malformations and intellectual disability [67, 68]. Unravelling the roles of these factors in NMD could provide valuable insights into fundamental gene regulatory mechanisms and offer potential avenues for therapeutic intervention in rare genetic disorders.

To elucidate the cellular functions of SMG8 and SMG9, we investigated their proposed role as negative regulators of SMG1-mediated phosphorylation of UPF1. This role is largely based on in vitro studies demonstrating that SMG8, via its C-terminal KID, inhibits SMG1 kinase activity, while SMG9 is required to stabilize the SMG8–SMG1 interaction [16, 31–33, 35]. These findings support a model in which SMG1 exists in an autoinhibited state that is released upon engagement with NMD targets. To test whether this mechanism also applies in vivo, we analysed UPF1 phosphorylation in HCT116 cells lacking the SMG8 KID or carrying knockouts of SMG8 or SMG9. None of these perturbations resulted in pronounced hyperphosphorylation of UPF1 under steady-state conditions. Instead, loss of SMG8 or SMG9 caused modest and variable increases in UPF1 phosphorylation that were detectable in several, but not all, cellular contexts and clones. These increases were modest compared with expectations from in vitro studies, but compatible with earlier in vivo observations [35]. Together, our findings argue against a model in which SMG8 and SMG9 act as constitutive inhibitors of SMG1 activity in cells. Instead, they support a role for SMG8 and SMG9 as modulatory components that differentially influence SMG1 activity, with SMG9 having a stronger effect on UPF1 phosphorylation in vivo. This model aligns with available structural data, which position SMG9 as a bridging component between SMG8 and SMG1 [30, 32], potentially explaining why SMG9 loss has a more pronounced effect on UPF1 phosphorylation than SMG8 loss in cellular systems. We note that our immunoblot-based assays primarily capture global steady-state phosphorylation and may not fully resolve localized, transient, or transcript-specific phosphorylation events.

Recent findings support a more dynamic and context-dependent model of UPF1 phosphorylation, in which its modification is not solely dictated by the SMG1:SMG8:SMG9 complex but also shaped by the progression and efficiency of downstream NMD events [18]. In this model, UPF1 becomes hyperphosphorylated when degradation is delayed or stalled, thereby extending its interaction with decay-promoting factors and amplifying subsequent steps in the pathway. Accordingly, the phosphorylation state of UPF1 reflects not only SMG1 kinase activity but also the residence time of UPF1 on its RNA targets and the effective execution of decay. If, in the absence of SMG8 or SMG9, SMG1 activity becomes less selective, phosphorylating UPF1 bound to both normal and premature termination codon (PTC)-containing transcripts, this could result in a redistribution of phosphorylation events without significantly altering the total pool of phosphorylated UPF1. As our assays measure overall phosphorylation levels, such shifts in substrate specificity would remain undetected. This context-dependent regulation offers a plausible explanation for why we did not observe an increase in UPF1 phosphorylation in SMG8- or SMG9-deficient cells.

Unlike SMG8 and SMG9, SMG1 is essential for viability in cultured human cells, precluding genetic knockout approaches. To circumvent this, we used the selective SMG1 inhibitor SMG1i, which effectively blocks SMG1 activity, reduces UPF1 phosphorylation, and inhibits NMD. SMG1i was originally developed as a targeted anti-SMG1 compound with potential therapeutic relevance in oncology [49]. Its mechanism of action aligns with the role of NMD in immune evasion, where the degradation of PTC-containing transcripts suppresses tumour-specific neoantigen production. Inhibiting NMD, either through SMG1i or the alternative SMG1 inhibitor KVS0001, can enhance neoantigen expression and promote T cell-dependent anti-tumor immunity [69, 70]. Accordingly, NMD inhibition has been shown to suppress tumour growth in vivo [71, 72]. Our data demonstrate that SMG8- and SMG9-deficient cells are markedly more sensitive to SMG1i than wild-type cells. Such NMD-sensitized backgrounds offer a powerful tool for functional dissection of the pathway, as they reveal the roles of individual factors that may be invisible in a fully intact system. In this context, SMG1 inhibition serves as a targeted perturbation that enables the identification of nonessential NMD components. Importantly, these effects were not restricted to a single cellular context but were reproducibly observed across multiple human cell lines. This consistency indicates that the sensitization of NMD to partial SMG1 inhibition upon loss of SMG8 or SMG9 reflects a general property of NMD rather than a cell line-specific phenomenon.

In the context of the previously proposed two-step authentication model, our data provide further insight into the intricate architecture of NMD regulation. In this model, phosphorylation of UPF1 by the SMG1:SMG8:SMG9 complex represents the first authentication step. This enables UPF1 to recruit the SMG5:SMG7 complex, which then activates SMG6-mediated endonucleolytic cleavage, constituting the second step. Although UPF1 phosphorylation appears largely intact in SMG8- or SMG9-deficient cells, the first authentication step is partially compromised, making cells more susceptible to additional perturbations (Fig. 6C). Similar cooperativity between the two steps is evident when SMG1i is combined with SMG6 knockdown, SMG7 knockout, or CASC3 knockout. Each manipulation alone leads to partial NMD inhibition, but in combination, results in complete loss of NMD function. This observation can be used to validate the role of factors in NMD in cases where their absence leads to only partial inhibition. Notably, SMG1i would not be expected to synergize with NMD-independent effects, but would be specific to perturbations of the NMD pathway.

Taken together, our results support a model in which mammalian NMD is a resilient, multilayered system capable of withstanding partial disruption through the presence of redundant and compensatory factors. These include not only SMG8 and SMG9, but also paralogous proteins such as UPF3A and UPF3B, and duplicated EJC components like MAGOH and MAGOHB (Fig. 6C). Lack of any of these factors does not immediately result in dramatic consequences for NMD. Instead, it merely reduces its capacity and error tolerance, which, in turn, only manifests as phenotypically observable effects under particularly challenging conditions. This may explain why mutations in SMG8 and SMG9 are associated with variable developmental phenotypes rather than universal NMD inhibition [67, 68]. It is plausible that only certain tissues at certain times of development require a particularly robust NMD to develop normally, making them more sensitive to subtle defects in NMD pathway regulation. Hence, our findings not only provide mechanistic insight into how NMD activity is tuned in vertebrates but also offer a foundation for future therapeutic strategies targeting NMD dysfunction in disease.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work was supported by grants from the Deutsche Forschungsgemeinschaft (DFG, CRC 1678, grant agreement no. 520471345) and the Center for Molecular Medicine Cologne (CMMC) (project C05) to N.H.G. Additionally, this work was supported by the DFG Research Infrastructure as part of the Next Generation Sequencing Competence Network (project 423957469) and by the large instrument grant INST 216/1163-1 FUGG from the DFG (DFG Großgeräteantrag). Funding to pay the Open Access publication charges for this article was provided by the CRC 1678 (grant agreement no. 520471345).

Data availability

This study analyzes publicly available data, which are listed in Supplementary Table S1.

RNA-seq data generated in this study have been deposited at BioStudies/ArrayExpress and are publicly available as of the date of publication (BioStudies accession E-MTAB-13949 and E-MTAB-16399).

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE [73] partner repository with the dataset identifier PXD051058.

The resource data underlying this article including raw image blots, quantifications and high-throughput analysis results are available at Zenodo (https://doi.org/10.5281/zenodo.18299536).