UPF1 mutants with intact ATPase but deficient helicase activities promote efficient nonsense-mediated mRNA decay

Abstract

The conserved RNA helicase UPF1 coordinates nonsense-mediated mRNA decay (NMD) by engaging with mRNAs, RNA decay machinery and the terminating ribosome. UPF1 ATPase activity is implicated in mRNA target discrimination and completion of decay, but the mechanisms through which UPF1 enzymatic activities such as helicase, translocase, RNP remodeling, and ATPase-stimulated dissociation influence NMD remain poorly defined. Using high-throughput biochemical assays to quantify UPF1 enzymatic activities, we show that UPF1 is only moderately processive (<200 nt) in physiological contexts and undergoes ATPase-stimulated dissociation from RNA. We combine an in silico screen with these assays to identify and characterize known and novel UPF1 mutants with altered helicase, ATPase, and RNA binding properties. We find that UPF1 mutants with substantially impaired processivity (E797R, G619K/A546H), faster (G619K) or slower (K547P, E797R, G619K/A546H) unwinding rates, and/or reduced mechanochemical coupling (i.e. the ability to harness ATP hydrolysis for work; K547P, R549S, G619K, G619K/A546H) can still support efficient NMD of well-characterized targets in human cells. These data are consistent with a central role for UPF1 ATPase activity in driving cycles of RNA binding and dissociation to ensure accurate NMD target selection.

INTRODUCTION

The nonsense-mediated mRNA decay (NMD) pathway is a translation-dependent mechanism responsible for both quality control of mRNAs with premature termination codons (PTCs) and general gene expression regulation (1,2). The RNA helicase UPF1 is the lynchpin of NMD, serving as a protein scaffold for decay complex assembly (3). Several components of the NMD pathway directly interact with UPF1 to promote decay, including UPF2 (4–6), the PI3K-like kinase SMG1 (7,8), the SMG5/SMG7 heterodimer (9–11), the endonuclease SMG6 (9–13) and the terminating ribosome (14,15). In metazoans, phosphorylation of RNA-bound UPF1 by SMG1 is an important pro-decay signaling event, as it promotes assembly of decay enzymes on the mRNA (11–13,16). These observations suggest that prolonged association of UPF1 with an mRNA is a prerequisite for NMD, as it permits phosphorylation by SMG1, sensing of translation termination, and assembly of RNases on the mRNA.

Mutations in UPF1 that disrupt ATP binding or hydrolysis enable more stable UPF1-RNA binding in the presence of ATP (17,18) and result in hyperphosphorylated UPF1 in mammalian cells (19). As the accumulation of phospho-UPF1 in mRNPs is thought to be a strong signal for RNA decay, these features alone might be expected to promote NMD. However, ATPase-deficient mutants strongly impair NMD in yeast (14,15,17,20,21) and human (22,23) cells, suggesting that UPF1 ATPase activity is important for one or more steps in NMD after UPF1 associates with the mRNA. As a superfamily 1 (SF1) helicase, UPF1 uses the energy from ATP hydrolysis to generate work in the form of 5′-3′ translocation on nucleic acids (18,24–27), duplex unwinding, and disruption of RNA-protein interactions (28,29). These activities are proposed to aid in recycling ribosomes at PTCs (14,15), displace RNA-stabilizing factors prior to decay initiation (28) and disassemble NMD complexes to allow complete mRNA degradation by XRN1 (23,30), a processive 5′-3′ exonuclease (31–34). Models to explain the requirement for UPF1 enzymatic activity thus span the entire process of decay, from initial substrate selection (35) to post-decay messenger ribonucleoprotein (mRNP) disassembly (14,15,23). However, experimental systems capable of distinguishing among specific UPF1 enzymatic features (including ATPase, helicase, translocase, RNP remodeling, and ATPase-stimulated dissociation) and their roles in decay are lacking, limiting mechanistic understanding of the enzymatic roles of UPF1 in NMD.

Here, we develop an experimental framework for (i) quantitative measurement of UPF1 enzymatic properties in vitro and (ii) evaluation of the NMD functionality of UPF1 mutants with alterations in those enzymatic properties in human cells. First, using ensemble and single-molecule approaches, we find that UPF1 has only moderate processivity (10s–100s of nucleotides [nt]) when the nucleic acid substrate is not subjected to external forces that destabilize the duplex (e.g. as in magnetic tweezers experiments). We have developed an array of novel high-throughput biochemical assays to quantify the enzymatic properties of UPF1 and computationally identified UPF1 mutants with altered RNA binding properties. Combining these approaches, we quantify the enzymatic properties of a panel of UPF1 ATPase/helicase mutants. We further find that UPF1 mutants with poor processivity, slow unwinding rate, impaired protein displacement activity and/or impaired coupling between ATPase activity and unwinding/translocation efficiently restore NMD upon depletion of endogenous UPF1 in human cells. These data are consistent with a role for UPF1 ATPase activity in sampling potential substrates (35), but call into question whether efficient mechanochemical coupling and processive RNP remodeling of UPF1 are necessary to conduct NMD.

MATERIALS AND METHODS

Oligonucleotides and chemicals

All oligonucleotides were purchased from IDT and resuspended in TE. Biotin–BSA (Thermo Scientific 29130) was dissolved to 2 mg/ml in water, aliquoted and stored at −20°C. NeutrAvidin (Thermo Scientific 31002) was dissolved to 5 mg/ml in water, aliquoted and stored at −20°C. Alexa Fluor 647-labeled Streptavidin (Invitrogen) was dissolved to 2 mg/ml in PBS, aliquoted and stored at −20°C protected from light. Anti-Digoxigenin (Roche 11333089001) was dissolved to 0.2 mg/ml in PBS, aliquoted and stored at −20°C. Doxycycline (Sigma D9891) was dissolved to 1 mg/ml in water and stored at 4°C protected from light. ATP, AMP-PNP, ATPγS, and ADP were all purchased from Millipore Sigma. ATP was diluted in 10 mM MES pH 6.0 and stored at −80°C, except for use in equilibrium binding assays and FAD assays involving comparison to other nucleotide analogs, in which case it was diluted in 10 mM Tris–HCl pH 7.5 and stored at −80°C, as with AMP-PNP, ATPγS, and ADP. All nucleotide analogs were diluted to 5 mM in 10 mM MES pH 6.0 for kinetic assays.

UPF1-HD protein expression and purification

Protein expression and FPLC purification of UPF1-HD protein with N-terminal 6xHis and calmodulin binding peptide tag was performed as previously described (29). For batch purification, after 2–3 passes through a French Press at 1000 psi, extracts were filtered through a 0.44 μm filter, rotated at 4°C for 1.5–2 h in 0.75 ml equilibrated Ni-NTA Agarose (Qiagen), followed by packing pre-equilibrated Poly-Prep Chromatography Columns (Bio-Rad). Columns were washed with 10 ml binding buffer (1.5× PBS pH 7.5, 225 mM NaCl, 1 mM MgOAc, 0.1% NP-40, 10% glycerol) supplemented with 20 mM imidazole, followed by an additional 2.5 ml binding buffer supplemented with 50 mM imidazole, and proteins were eluted in 500–600 μl binding buffer supplemented with 500 mM imidazole. Dialysis and flash freezing was done as previously described (29). UPF1-HD was diluted in protein dilution buffer (100 mM NaPO₄ pH 7.2, 150 mM NaCl) for day-of use.

In vitro transcription and substrate annealing for unwinding, dissociation, and PIFE assays

To generate RNA substrates, PCR using custom primers (Supplementary Table S1), in vitro transcription, and quantification were performed as previously described (29), except using T7 RNA Polymerase and pyrophosphatase generously provided by the Ferré-D’Amaré lab. Annealing of oligonucleotide substrates (Supplementary Table S1) was also performed as previously described in 500 μl total volume (29). Where indicated, a 1:1 substrate strand to fluorescent strand ratio was used in lieu of the standard 11:7 ratio. Annealed substrates were electrophoresed on native Novex TBE Gels (ThermoFisher Scientific) to confirm the absence of unannealed fluorophore strands.

Fluorescence-based unwinding assays

Multiple turnover and single turnover unwinding assays were performed as previously described (29). Briefly, 82.5 nM UPF1-HD was pre-bound to 75 nM fluorescent substrate, and 75 μM unlabeled trap strand and 562.5 nM quencher strand were added immediately before loading into the plate reader. 16 μl of 5 mM ATP (2 mM final) was injected into each well and fluorescence of the 40 μl reaction was immediately measured over time. For unwinding assays performed at pH 7.5, 20 mM Tris–HCl pH 7.5, 75 mM KOAc, 3 mM MgCl₂ and 1 mM freshly prepared DTT were used. Measurements were performed using Tecan Spark or CLARIOstar Plus plate readers (BMG Labtech). As a single turnover control, 75 μM unlabeled trap strand was added before UPF1-HD addition. Unlabeled trap strand was omitted in multiple turnover unwinding reactions. Details of processivity calculations and simulations and unwinding data fitting are provided in Supplementary Materials and Methods.

Magnetic tweezers substrate generation

DNA hairpin substrates were created as previously described using custom oligonucleotides (Supplementary Table S1) (36). The surface-proximal handle was annealed to a 75 nt strand, and the bead-proximal handle was annealed to a 50 nt strand to create a 25 nt single-stranded UPF1-HD loading region immediately 5′ of the junction. The hairpin body was created via PCR of a 495 bp fragment of lambda DNA to generate a 535 bp hairpin stem.

Magnetic tweezers force-based unwinding assay

Experiments were performed as previously described (36). Briefly, a 30 μl reaction containing 50 pM hpDNA substrate and 6.67 ng/μl anti-digoxigenin was added to a KOH-cleaned and assembled flow cell with polystyrene beads melted on the surface (36). After incubating at 4°C overnight, 200 μl wash buffer (PBS supplemented with 0.3% BSA and 0.02% Tween-20) was flowed through the flow cell, which was then placed on a PicoTwist magnetic tweezers apparatus and allowed to settle for 30 min. Next, 5 μl of 10 mg/ml Dynabeads MyOne Streptavidin T1 magnetic beads (Invitrogen 65601) were washed twice in 200 μl bead wash buffer (10 mM Tris–HCl pH 7.0, 2 M NaCl, 1 mM EDTA), resuspended in 200 μl wash buffer, and sonicated for 5 min, after which 40 μl was flowed into the flow cell and incubated for 5 min. Next, 1–2 ml wash buffer was flowed through until most unbound beads were gone, after which 0.5–1 ml UPF1 wash buffer (20 mM Tris–HCl pH 7.5, 75 mM KOAc, 3 mM MgCl₂, 1 mM DTT, 1% BSA, 0.02% Tween-20) was flowed through, and tethers were located and calibrated. Only tethers that extended 0.4–0.5 μm as the force was increased from 7–9 pN to 20–22 pN were used in experiments. 200 μl UPF1 wash buffer supplemented with 20 nM UPF1-HD was then passed through 3 times, waiting 3–4 min after each addition. A final 200 μl was passed through, incubated for 10 min, and 100 μl UPF1 wash buffer supplemented with 2 mM ATP was added to the flow cell, 25–50 μl of which was passed through to initiate UPF1-HD unwinding.

Tether unwinding substrate generation

After annealing four oligonucleotides (hairpin strand, digoxigenin strand, biotin strand, Alexa Fluor 488 strand, Supplementary Table S1) in an 80 μl reaction, 7.5 μl T4 DNA Ligase in T4 DNA Ligase Buffer (NEB) was added, and the reaction was incubated at room temperature for 1 h, followed by heating at 65°C for 10 min and cooling to room temperature for 1 h on a tube block. Reactions were split in two and passed through two CHROMA SPIN + TE-100 columns (Takara 636072), following the manufacturer's protocol. Denaturing (Novex TBE-Urea) and native (Novex TBE) gels were run for quantification and quality checks.

Tether unwinding assay

Biotin–BSA was dissolved to 1 mg/ml in tether buffer (50 mM Tris–HCl pH 7.5, 50 mM NaCl). 5 μl was added to the center of a well in a 24-well glass bottom plate (Cellvis P24-1.5H-N) and incubated for 15 min. After three washes with 5 μl tether buffer, 5 μl of 1 mg/ml NeutrAvidin (in tether buffer) was added and incubated for 15 min to bind exposed biotin moieties. After three tether buffer washes, the 5 μl spot was washed 3 times with passivation buffer (2% BSA, 1 mM DTT, and 2 mM MgOAc). Next, 200 nM hairpin substrate pre-incubated with 200 nM UPF1-HD at room temperature in passivation buffer for 10–15 min was added to the spot, incubated for 15 min, and washed 3 times with passivation buffer to remove unbound substrates. Lastly, 15 μl passivation buffer supplemented with 2 mM ATP and 600 nM of a 38 nt quencher strand was added and Alexa Fluor 488 fluorescence was measured over time, as in solution fluorescence-based unwinding assays at 37°C in a CLARIOstar Plus plate reader (BMG Labtech).

Single-molecule picometer-resolution nanopore tweezers (SPRNT) substrate generation

All DNA oligonucleotides for SPRNT experiments were ordered from panoligo at Stanford University. Hairpins were formed by rapid annealing, in which 1 μM DNA was heated to 90°C then cooled to 4°C over 4 min.

SPRNT unwinding and translocation assays

SPRNT assays were performed as previously described (37). 4.5 μM UPF1-HD was incubated with 1 μM DNA at 37°C, of which 1 μl was added to a 100 μl reaction volume. Experiments were run at 1 mM or 2.5 mM ATP supplemented with 10 mM MgCl₂ and 10 mM DTT. MspA nanopores insert in a random orientation, and because MspA is a non-ohmic pore, the pore orientation is determined by measuring the current-voltage response of the pore. At 500 mM KCl, a forwards pore has a current of 180 ± 10 pA at 180 mV and −220 ± 10 pA at −180 mV. Conversely, a backwards pore has a current of 220 ± 10 pA at 180 mV and −180 ± 10 pA at −180 mV. Details of processivity analysis from SPRNT measurements are provided in Supplementary Materials and Methods.

Fluorescence anisotropy dissociation (FAD) assay

15 μl reaction buffer (10 mM MES pH 6.0, 50 mM KOAc, 0.1 mM EDTA, 2 mM MgOAc, 2 mM DTT) supplemented with 14.25 nM substrate (Supplementary Table S1) was added to a 96-well half-area black microplate (Corning 3993), followed by addition of 3 μl of 950 nM UPF1-HD (71.25 nM final) and incubation for 10 min. Next, 3 μl of 190 μM trap strand was added (14.25 μM final), and the plate was loaded into a Tecan Spark or CLARIOstar Plus plate reader (BMG Labtech) pre-heated to 37°C. Fluorescence polarization of Alexa Fluor 488 was measured over time, beginning immediately after injection of 16 μl of 5 mM ATP (2 mM final) into wells. To obtain a baseline of unbound substrates, 14.25 μM trap strand was added prior to UPF1-HD addition, and constant low polarization over time was observed because very few fluorescent strands were protein-bound. Five-fold excess UPF1-HD was used in order to increase the dynamic range of the assay. Half-life was computed by extracting the time at which fluorescence polarization decreased halfway to the trap control baseline.

UPF1-HD fluorescent labeling with cysteine-maleimide chemistry

Batch purified UPF1-HD was re-dialyzed in 500 ml DTT-free buffer (1.5× PBS (KD Medical), 125 mM NaCl, 0.66 mM MgOAc, 1 μM ZnSO₄) in 0.5 ml Slide-A-Lyzer Dialysis Cassettes (Thermo Scientific 66454) at 4°C for 4 h replacing buffer every hour. 1 mM TCEP was added to reduce cysteine residues for 10 min on ice. Next, 100 μM Alexa Fluor 647 C₂ Maleimide (10-fold molar excess, Thermo Fisher Scientific) was added and rotated at 4°C overnight. 10 ml DTT-free dialysis buffer and 10 ml regular dialysis buffer were freshly made and cooled on ice. Zeba Spin Desalting Columns (Thermo Fisher Scientific 89882) were equilibrated following the manufacturer's protocol. Samples were passed through 3 times in columns equilibrated with DTT-free dialysis buffer, then passed through once in columns equilibrated with regular dialysis buffer.

Simultaneous unwinding/dissociation (SUD) assays

SUD experiments were performed in a similar manner as single turnover unwinding assays, except 250-fold excess of dual quencher (12 nt oligonucleotide dual labeled with Iowa Black RQ to quench UPF1-HD-647, Supplementary Table S1) strand was substituted for the 1000-fold excess of unlabeled trap strand, and UPF1-HD-647 for UPF1-HD. All experiments were performed on a CLARIOstar Plus plate reader at 37°C (BMG Labtech).

NADH-coupled ATPase assays

Assays were performed as previously described where 80 nM UPF1-HD was pre-bound to 1 μM substrate (29) but (Supplementary Table S1) supplemented with 1 mM DTT and using clear 96-well half area plates (Greiner Bio-one 675101). All experiments were performed on a CLARIOstar Plus plate reader at 37°C with 1 mM ATP (BMG Labtech). Details of ATPase assay data analysis are provided in Supplementary Material & Methods.

Fluorescence anisotropy binding assays

37 ul master mix (20 mM Tris–HCl pH 7.5, 75 mM KOAc, 3 mM MgCl₂, 1 mM freshly prepared DTT, 1 mM nucleotide, and indicated concentration of Alexa Fluor 488 labeled DNA, Supplementary Table S1) was added to a 96-well half area black plate (Corning 3993) followed by addition of 3 ul protein diluted in protein dilution buffer (100 mM NaPO₄ pH 7.2, 150 mM NaCl) and incubated for the indicated time points. Wells were measured using the fluorescence polarization module on the Tecan Spark plate reader. Note that the apparent binding affinities observed here may be underestimated, as substrate concentrations <0.5 nM were below the limit of detection for this assay.

Protein induced fluorescence enhancement (PIFE) assays

18 μl reaction buffer (10 mM MES pH 6.0, 50 mM KOAc, 0.1 mM EDTA, 2 mM MgOAc, 2 mM DTT) supplemented with 75 nM DNA or 38 nM RNA substrate (Supplementary Table S1) was added to a 96-well half-area black microplate (Corning 3993), followed by addition of 3 μl of 1.1 μM or 550 nM UPF1-HD (82.5 or 41.25 nM final) and incubation for 10 min. Next, 3 μl of 1 mM or 0.5 mM trap strand was added (75 or 37.5 μM final), and the plate was loaded into a CLARIOstar Plus plate reader (BMG Labtech) pre-heated to 37°C. Cy3 fluorescence intensity was measured over time, starting immediately after injection of 16 μl of 5 mM ATP (2 mM final) into wells. To obtain a baseline of unbound substrates, 75 or 37.5 μM trap strand was added prior to UPF1-HD addition. Half-life was computed by extracting the time at which fluorescence polarization decreased halfway to the trap control baseline.

Fluorescence-based biotin-streptavidin displacement assays

15 μl reaction buffer (10 mM MES pH 6.0, 50 mM KOAc, 0.1 mM EDTA, 2 mM MgOAc, 2 mM DTT) supplemented with 75 nM of a 3′ biotin-labeled DNA substrate (Supplementary Table S1) was added to a 96 well half-area black microplate (Corning 3993), followed by addition of 3 μl of 2.2 μM UPF1-HD (165 nM) and incubation for 5 min. Next, 3 μl of 200 nM Alexa Fluor 647-labeled Streptavidin tetramer (Invitrogen, 15 nM final) was incubated in the reaction for 10 min, and 3 μl of 63 μM of a 12 bp biotin-quencher was added (4.725 μM final). The biotin-quencher was generated by annealing a 12 nt 5′ biotin-labeled oligonucleotide with dual quencher (see SUD assay), following the standard protocol but using 1.23-fold excess of dual quencher. After a 10 min incubation to allow the reaction to come to equilibrium, the plate was loaded into a CLARIOstar Plus plate reader (BMG Labtech) pre-heated to 37°C. Alexa Fluor 647 fluorescence intensity was measured over time, starting immediately after injection of 16 μl of 5 mM ATP into wells (2 mM final). To obtain a baseline of quenched streptavidin, 4.725 μM biotin-quencher was added prior to Alexa Fluor 647-labeled Streptavidin addition.

In silico UPF1-HD structure preparation for mutational screen

The UPF1-HD crystal structure (PDB 2XZO (38)) was modified before performing the in silico screen. First, the terminal nucleotide was patched using rosetta/2019.42/main/database/chemical/residue_type_sets/fa_standard/patches/nucleic/rna/3prime_phosphate.txt and subsequently minimized. The structure was further curated with BioLuminate (https://www.schrodinger.com/products/bioluminate) using the Protein Preparation Wizard tool. Missing side chains were filled in using Prime, and missing loops were filled in by referencing a curated database of known loops in the PDB. The curated structure was relaxed 100 times using rosetta/2019.42/main/source/src/apps/public/rnp_ddg/get_lowest_scoring_relaxed_models.py–relax_dir relax_reprocessed > lowest_scoring_relaxed_structures.txt, where the 20 lowest ΔΔG structures were used for downstream analysis. This structure is in a post-hydrolysis state, which maximizes the likelihood of finding important interactions with RNA at an intermediate state in the ATPase/helicase cycle.

In silico Rosetta-Vienna RNP ΔΔG method (39) for mutational screen

Sequences of every possible point mutation were generated using a custom Python script and subsequently split into 100 text files to parallelize computing. Custom bash scripts were run on the NIH high-performance computing system, Biowulf, to complete the Rosetta-Vienna RNP ΔΔG method (39,40). Each mutant was run 20 times to calculate the average and standard deviation ΔΔG values.

Site-directed mutagenesis of UPF1-HD plasmids for bacterial expression

The pET28 vector harboring UPF1-HD with an N-terminal 6xHis and calmodulin binding peptide tag was used as the template. Mutations were generated by following the Q5 site-directed mutagenesis kit protocol (NEB E0554) with Q5 Hot-Start High-Fidelity Master Mix, 10 ng template plasmid, and 500 nM of appropriate primers (designed via the NEBaseChanger website, Supplementary Table S1). PCR was initiated at 98°C for 30 s, followed by 25 cycles of 98°C 10 s, T_a 20 s, 72°C 3.5 min, and a final 72°C extension for 2 min. An aliquot of the 25 μl reaction was run on a 1% ethidium bromide agarose gel (or 1% E-Gel agarose gel, Thermo Fisher Scientific) to verify one unique PCR product band. The PCR reaction was purified using the Qiagen QIAquick PCR Purification Kit eluting in 30 μl buffer EB. 1 μl was used in a 10 μl Kinase/Ligase/DpnI (KLD, NEB) reaction and incubated at room temperature for 10–20 min. Following KLD treatment, 5 μl was transformed into either NEB 5-alpha competent Escherichia coli cells or homemade DH5-alpha competent cells and plated onto LB agar plates with 30 μg/ml kanamycin (Teknova L1024). After a 37°C overnight incubation, colonies were picked into 3 ml LB supplemented with 50 μg/ml kanamycin and, after a 37°C 230 rpm incubation overnight, cultures were subjected to miniprep using Qiagen QIAprep Spin Miniprep Kit eluting in 30 μl buffer EB. Clones were sequence verified with custom primers (Supplementary Table S1, Psomagen).

Site-directed mutagenesis of CLIP-UPF1 plasmids for mammalian overexpression

The pcDNA5-FRT-TO vector harboring N-terminal CLIP-tagged full length UPF1 was used as the template (41). Mutations were generated as above, but due to the high GC content and vector size, several modifications were necessary. Phusion high-fidelity DNA polymerase in Phusion GC buffer and 3% DMSO were used to set up PCR reactions, which were initiated at 98°C for 5 min followed by 25 cycles of 98°C 1 min, T_a (minus 5–6°C 30 s, 72°C 5 min, and a final 72°C extension for 10 min. If multiple bands appeared on the gel, the entire PCR reaction was rerun on the gel, bands were cut out, and purified using the Qiagen MinElute Gel Extraction Kit eluting in 10 μl buffer EB. Transformed competent cells were plated onto LB agar plates with 100 μg/ml carbenicillin (Teknova L1010), and colonies were picked into LB supplemented with 50 μg/ml carbenicillin. For transfection-grade plasmids, the NucleoSpin Miniprep kit was used (Macherey-Nagel 740490), and plasmids were eluted in 50–60 μl elution buffer. Clones were sequence verified with custom primers (Supplementary Table S1, Psomagen).

Cloning of CMV (−40)+1 CLIP-UPF1 plasmids for mammalian expression

The pcDNA5-FRT-TO-CLIP-UPF1 plasmids are overexpressed in mammalian cells, so a −40 kcal/mol hairpin sequence was placed 1 nt downstream of the predicted transcription start site to lower CLIP-UPF1 protein expression. The tetracycline-regulated promoter of the pcDNA5-FRT-TO sequence was replaced with the CMV promoter and hairpin sequence by cut-and-paste cloning. 1 μg of each pFRT-TO-CLIP-UPF1 mutant vector was digested with 1 μl SpeI (NEB) and 1 μl HindIII-HF (NEB) in CutSmart buffer in 20 μl at 37°C overnight. The same reaction was performed to cut out the CMV (−40)+1 sequence but with 3 μg plasmid. Restriction digests were run on 1% E-Gel EX agarose gels (Thermo Fisher Scientific), and bands were cut out. Vector digest bands were purified using the Qiagen QIAquick Gel Extraction Kit eluting in 30 μl buffer EB, whereas the CMV (−40)+1 insert sequence band was purified using the Qiagen MinElute Gel Extraction Kit eluting in 10 μl buffer EB. A 10 μl reaction containing 3 μl vector, 3 μl insert, and 1 μl T4 DNA Ligase (NEB) in T4 DNA Ligase buffer was incubated at room temperature for 30–60 min, and subsequently transformed into DH5-alpha cells and cultures were grown, purified, and sequence verified as above.

Mammalian stable cell line generation

T-Rex-293 cells (Invitrogen) were cultured and maintained as previously described (41) in DMEM (Gibco 11-965-092) supplemented with 10% FBS (Gibco) and 1x penicillin/streptomycin/l-glutamine (Gibco 10378016) at 37°C and 5% CO₂. To generate stable lines, 500,000 cells were split into each well in a 6-well plate (Corning 353046). 16–18 hr later, cells were transfected using the Lipofectamine 3000 Reagent (Thermo Fisher Scientific) with 200 ng pcDNA5-FRT-TO-CLIP-UPF1 and 2.25 μg pOG44 Flp-Recombinase Expression Vector (or pcDNA3.1 as a control). Media was replaced 6–7 h post-transfection. Cells were transferred into 10 cm plates (Corning 353003) 2–3 days post-transfection, and media was replaced after 4–16 h with media supplemented with 100 μg/ml Hygromycin B (Invitrogen) to start selection. Once visible colonies started forming, cells were transferred into 15 cm plates (Corning 353025). After 1–2 splits, cells were trypsinized (Gibco 12605), transferred into 15 ml tubes, centrifuged at 1000 × g for 1 min, washed in DPBS (Gibco 14190), and resuspended in 5 ml freeze media (DMEM supplemented with 30% FBS and 10% DMSO), then aliquoted into cryogenic vials, slowly frozen at −80°C in CoolCell Alcohol-Free Cell Freezing Containers (Thomas Scientific) for at least 3 days, then transferred onto dry ice then into liquid N₂ tanks for long-term storage.

Protein extraction from cell lines and CLIP tag labeling

Cells in 12-well plates (Corning 353043) were lysed in 300 μl Passive Lysis Buffer (Promega) and centrifuged at maximum speed for 1–2 min at 4°C. The supernatant was transferred to pre-cooled tubes, of which 10 μl was mixed with 150 μl Pierce 660 nm Protein Assay Reagent (Thermo Scientific) in a clear 96-well plate (Greiner Bio-one 655101). After a 5 min incubation on a rocker, absorbance at 562 nm was measured on a Tecan Infinite M200 Pro plate reader. Following quantification, a 25 μl reaction containing 0.2 mg/ml extract, 1 mM DTT, and 8 μM BC-647 (CLIP-Surface 647, NEB) in a base solution of Passive Lysis Buffer was rotated overnight at 4°C to label CLIP-tagged proteins. An aliquot of this reaction was mixed with 8 mM DTT and NuPAGE LDS Sample Buffer (Invitrogen), heated at 70°C for 5 min, then loaded on a NuPAGE 4–12% Bis–Tris Mini Protein Gel in MOPS buffer with 500 μl NuPAGE Antioxidant (Invitrogen) in the inner chamber. The gel was run at 100 V for 10 min then 200 V for 40 min, imaged using an Amersham Typhoon (GE), and quantified using Fiji.

siRNA treatment and cell harvesting

6 μl of 20 μM siRNA (siUPF1: sense = CUACCAGUACCAGAACAUAtt, antisense = UAUGUUCUGGUACUGGUAGgc, siNT: AN2) were added to six-well plates, then 1.5 ml master mix (1.5 ml Opti-MEM plus 5.25 μl Lipofectamine RNAiMAX) was added and incubated for 30–50 min. During the incubation, cells were trypsinized, counted, and diluted to 2 × 10⁵ cells/ml with media containing 20% FBS, of which 1.5 ml was added to each well. 24 h later, media was replaced with media supplemented with 200 ng/ml doxycycline to induce expression where applicable. 72 h post-siRNA transfection, cells were washed in DPBS and harvested by resuspending in 500 μl TRIzol Reagent (Invitrogen) and stored at −80°C.

Western blotting

Following cell lysis and quantification as above, extract was equilibrated with Passive Lysis Buffer (Promega), and aliquot was mixed with 8 mM DTT and NuPAGE LDS Sample Buffer (Invitrogen), heated at 70°C for 5 min, then loaded on a 1.0 mm NuPAGE 4–12% Bis–Tris Mini Protein Gel in MOPS buffer with 500 μl of NuPAGE Antioxidant (Invitrogen) in the inner chamber. The gel was run at 100 V for 10 min then 200 V for 40 min, then transferred to a 0.45 μm nitrocellulose membrane (BioRad 1620115) and transferred using the XCell II Blot Module (Invitrogen). Following transfer, membranes were cut, then incubated in 5 ml Blocking Buffer (Rockland MB-070) for 1 h at room temperature. Next, membranes were incubated overnight at 4°C with 5 ml primary antibodies diluted in Blocking Buffer. 1:5000 goat anti-Rent1 (Bethyl A300-038A) was used for UPF1 detection, and 1:5000 mouse anti-beta actin (Cell Signaling 3700) was used for beta actin detection. Next, membranes were washed 3 times in 5 ml TBS supplemented with 0.1% Tween-20, then incubated for 90 min at room temperature with 5 ml of 1:10 000 of either donkey anti-goat AlexaFluor 680 or goat anti-mouse AlexaFluor 680 secondary antibodies. Membranes were subsequently imaged using an Amersham Typhoon (GE) and quantified using Fiji.

RNA extraction, cDNA synthesis, RT-qPCR

RNA was extracted from mammalian cells by following the TRIzol Reagent manufacturer protocol but using 1 μl GlycoBlue (Invitrogen AM9515) to visualize nucleic acid pellets. DNase treatment using RQ1 DNase (Promega) was performed, and RNA was re-extracted using acid-phenol and precipitated with standard ethanol precipitation. DNase treatment using 1 μl Shrimp dsDNase (Thermo Scientific EN0771) in a 10 μl reaction for 5 min at 25°C then 55°C for 5 min was also used, followed by cleanup with 18 μl RNAclean XP beads following the manufacturer's protocol (Beckman Coulter). RNA concentration and purity were measured using a NanoDrop Spectrophotometer (Thermo Scientific). cDNA synthesis was performed using 500 ng RNA with Maxima H Minus cDNA Synthesis Master Mix (Thermo Scientific) following the manufacturer's protocol. The 10 μl cDNA reaction was diluted to 200 μl and 3 μl was used with iTaq Universal SYBR Green Supermix (Bio-Rad) and 670 nM of primer pairs (Supplementary Table S2) in a 10 μl reaction for immediate use in RT-qPCR. An epMotion 5073 (Eppendorf) was used to pipette cDNA, iTaq mix, and primers into a white 96-well semi-skirted plate (MIDSCI PR-PCR2196LC-W). The plate was sealed with Avant ThermalSeal optically clear polyester RT-PCR film (MIDSCI TS-RT2-100), briefly spun down, and loaded into a LightCycler 96 Instrument (Roche). Pre-incubation at 95°C for 90 s followed by 40 cycles of 95°C 10 s, 57°C 10 s, 72°C 10 sec was performed, taking fluorescence readings after the 72°C extension step was performed to obtain C_q values. Details of NMD efficiency calculations from RT-qPCR are provided in Supplementary Material & Methods.

RESULTS

In vitro unwinding assays reveal UPF1 processivity of less than 200 nucleotides (nt) in the absence of destabilizing forces

To better understand how the enzymatic activities of UPF1 impact NMD (Figure 1A), we first modified a recently reported real-time fluorescence-based unwinding assay (29,42) to measure UPF1 processivity and unwinding rates. We used the helicase core of UPF1 (UPF1-HD) to avoid auto-inhibition by its N-terminal CH (6,38,43) and C-terminal SQ (44) domains (Supplementary Figure S1A). In this assay, UPF1-HD is pre-bound to the fluorescent substrate, and upon addition of ATP, UPF1 either (i) completely unwinds and displaces the fluorescent strand, which then hybridizes to a complementary quencher strand, decreasing the fluorescence or (ii) dissociates before complete unwinding (Figure 1B, Supplementary Figure S1B, C). A lower fluorescence value at the end of the reaction therefore indicates more complete unwinding events. As controls, we observed no unwinding activity with ATPase-deficient DE636AA and K498A (17,18,21) UPF1-HD mutants (Supplementary Figure S1D, E).

Figure 1.

UPF1-HD processivity is less than 200 nt on RNA and DNA. (A) Schematic of UPF1 enzymatic activities on RNA and how they might translate into cellular activities. (B) Single turnover unwinding assay. Trap strand was omitted in multiple turnover assays. (C) Representative time-dependent fluorescence curves from single turnover unwinding measurements with varying lengths of 5′ ssRNA overhang (left). Trap control reactions included 1000-fold excess trap strand prior to UPF1-HD addition. Shaded areas represent standard deviation. Fraction of substrate unwound as a function of the 5′ ssRNA overhang length (right). Each colored dot represents a separate experiment from three independent experiments. Small dots represent technical replicates. Black lines indicate processivity model fits, and colored lines represent simulations of helicases with differing processivities (see Methods). (D, E) As in (C) but with varying lengths of 5′ ssDNA overhang (D) or hpDNA stem (E).

To account for different nucleic acid structures, we measured UPF1-HD translocation processivities on ssRNA and ssDNA, and unwinding processivities on dsRNA and hairpin DNA (hpDNA). In all cases, we found that UPF1-HD unwound a smaller proportion of substrate as the substrate length increased (Figure 1C–E, Supplementary Figure S1F, left graphs). We fit the data with established models (Materials and Methods) (45) to obtain the processivity for each type of substrate (Figure 1C–E, Supplementary Figure S1F, black lines on right graphs). We also compared the data to simulated helicases with differing processivities (Methods, Figure 1C-E, Supplementary Figure S1F, colored lines on right graphs). From these analyses, we determined UPF1-HD processivity to be 20–50 nt on ssRNA, ssDNA and dsRNA and 100–200 nt on hpDNA. Processivity is defined as a probability of forward translocation, but can be converted into the average number of steps taken before dissociation (Materials and Methods), which we refer to as processivity in the text for simplicity.

Surprisingly, these processivities are 2–3 orders of magnitude lower than those previously measured using magnetic tweezers, in which hairpin substrates were subjected to duplex-destabilizing forces (28,46,47). To investigate reasons for this discrepancy, we first performed additional controls to validate the fluorescence-based unwinding assay for processivity measurements. We first corroborated hpDNA processivity by performing an alternate processivity measurement based on a titration of UPF1-HD versus substrate concentration (Methods, Supplementary Figure S1G), and ensured that unwinding was primarily initiated at single-stranded regions (Supplementary Figure S2A). We next compared the pH 6.0 reaction buffer used in our experiments with the magnetic tweezers pH 7.5 reaction buffer under single (Figure 1E, left graph; Supplementary Figure S2B) and multiple turnover (Supplementary Figure S2C) unwinding conditions. This revealed substantially reduced unwinding activity at pH 7.5 compared to pH 6.0, consistent with the acidic preference of human UPF1 in vitro (27). We further verified that the concentration of the trap strand did not affect unwinding activity (Supplementary Figure S2D). Lastly, we altered the annealing ratios of substrate strand to fluorescent strand from 11:7 to 1:1 to ensure that the decrease in the fraction unwound on longer substrates was not due to longer unannealed strands acting as more effective UPF1-HD binding sinks (Supplementary Figure S2E, F). These control experiments reinforced our finding of significantly lower processivity of UPF1-HD in the absence of duplex-destabilizing force.

We next considered the possibility that, despite similar unwinding rates (Supplementary Figure S3A) (28,47,48), the UPF1-HD recombinant protein preparation used here was less processive compared to that used in magnetic tweezers experiments. Therefore, we performed unwinding assays on a 535 bp hpDNA under 7–9 pN of force using magnetic tweezers (Supplementary Figure S3B). In agreement with previous studies (28,46–48), we observed complete unwinding and translocation through the entire hairpin, indicating a processivity on the order of thousands of nt (Supplementary Figure S3C). To further evaluate the source of differential UPF1 processivity in ensemble and single-molecule assays, we performed ensemble unwinding experiments in which UPF1-HD unwound a fluorescent 70 bp hpDNA substrate tethered to a glass surface. In these experiments conducted without duplex-destabilizing force, UPF1-HD exhibited similar activity as in untethered fluorescence-based unwinding measurements (Supplementary Figure S3D, E). This suggests that the applied duplex-destabilizing force in magnetic tweezers, not tethering the substrate to the surface, is responsible for higher observed UPF1-HD processivity in magnetic tweezers. These observations are consistent with similar observations of force-dependent processivity enhancement of several RNA and DNA helicases (49–54).

Lastly, we investigated whether applying force on the nucleic acids in a non-duplex-destabilizing orientation would significantly increase UPF1-HD processivity. Therefore, we performed Single-molecule Picometer Resolution Nanopore Tweezers (SPRNT) unwinding and translocation measurements (37,55,56). In SPRNT, a single MspA nanopore in a phospholipid bilayer separates two electrolyte solutions. When a voltage is applied to the system, current flows through the nanopore, drawing the negatively charged DNA pre-bound to UPF1-HD into the nanopore. UPF1-HD controls the passage of the DNA through the nanopore, allowing measurement of UPF1-HD unwinding and translocation via the sequence-dependent changes in current caused by DNA moving through the nanopore (Figure 2A,B). Unlike magnetic tweezers, the applied force in SPRNT does not destabilize the duplex but instead assists UPF1-HD translocation by pulling the DNA through the nanopore.

Figure 2.

UPF1-HD exhibits low processivity in a single-molecule picometer-resolution nanopore tweezers (SPRNT) system. (A) Schematic of SPRNT system with a ‘forwards’ pore setup. In ‘backwards’ pore experiments, the MspA nanopore (yellow) is inverted 180° relative to the phospholipid bilayer. (B) Example unwinding event indicating the time-dependent ion current associated with the four stages of SPRNT signal illustrated in (A). The open pore current is high (1). The current drops when a UPF1-HD/DNA complex enters the pore (2). The current levels subsequently fluctuate as UPF1 unwinds the hairpin and the 5′ end of the DNA translocates through the pore (3). The current level rapidly rises when UPF1 dissociates and the DNA is pulled unimpeded through the pore (4). (C) Histogram of unwinding/translocation events and processivity fits for forwards (red) and backwards (blue) pore.

We used changes in current to temporally identify UPF1-HD-bound DNA initially docking on the nanopore, UPF1 unwinding and translocation, and subsequent UPF1-HD dissociation (Figure 2A, B). We aligned experimental traces (Supplementary Figure S4A) to the predicted current (Supplementary Figure S4B) to determine the position of UPF1-HD over time (Supplementary Figure S4C) and observed a processivity of ∼24 nt on hpDNA across multiple applied forces (Figure 2C, red). Since UPF1-HD interaction with the MspA nanopore may influence activity, we inverted MspA 180° relative to the membrane and observed similar processivity of ∼29 nt (Figure 2C, blue; Supplementary Figure S4D), but significantly higher unwinding and translocation rates, which increased even further with increasing assisting force (Supplementary Figure S4E, F). These results are consistent with fluorescence-based unwinding measurements of UPF1-HD processivity, and suggest that the non-duplex-destabilizing force in SPRNT does not enhance UPF1-HD processivity.

Taken together, our data suggest that UPF1 acts locally, associating and dissociating from cellular RNPs. Our findings are consistent with the hypothesis that the high processivity of UPF1-HD in magnetic tweezers experiments is due to force-induced processivity enhancement, as observed with other SF1(49–52), SF2 (54) and SF4 (53) helicases. Nevertheless, our results, together with the ability of UPF1 to generate sufficient force to disrupt biotin-streptavidin interactions (28,29), are compatible with a role for local mRNP remodeling by UPF1 for efficient NMD.

UPF1 undergoes frequent ATPase-stimulated dissociation events

Since the duration of UPF1 association with mRNA (residence time) has been hypothesized to be an important determinant of NMD specificity and efficiency (35), we next set out to measure UPF1-HD dissociation from nucleic acids. Although it is possible to estimate residence time from processivities and unwinding rates extracted from fluorescence-based unwinding assays, incomplete unwinding events may confound the results. Therefore, we developed a fluorescence anisotropy-based dissociation (FAD) assay to monitor UPF1-HD dissociation from fluorescent nucleic acids over time (Figure 3A). In this assay, each fluorescent substrate is pre-bound with ∼5 UPF1-HD molecules (Supplementary Figure S5A) to produce initial high fluorescence polarization, which subsequently decreases as UPF1-HD dissociates. UPF1-HD remained stably bound in the absence of ATP but underwent ATP-stimulated dissociation, with a half-life of ∼1 min on substrates with either 5′ or 3′ single-stranded overhangs (Figure 3B-C, Supplementary Figure S5B). In control measurements, ATPase-deficient DE636AA and K498A both remained stably bound in the presence and absence of ATP (Supplementary Figure S5C). If UPF1-HD processivity on ssDNA was high, a large population would translocate to the end of substrates before dissociating, and slower dissociation would be observed with longer substrates. However, we did not see significant differences in ATP-stimulated dissociation from substrates with 3′ overhangs over 40 nt (Figure 3D), confirming low processivity and indicating that the assay predominantly monitors UPF1-HD dissociation from internal sites.

Figure 3.

UPF1-HD undergoes ATP-stimulated dissociation from DNA and RNA. (A) Schematic of fluorescence anisotropy-based dissociation (FAD) assay. More arrows on fluorophore indicates faster tumbling and thus lower polarization. (B) Representative curves from FAD with a 3′ overhang substrate from four independent experiments (left). Trap control reactions included 1000-fold excess trap strand prior to UPF1-HD addition. Quantification of binding half-life (see Methods) from the time-dependent fluorescence polarization curves (right). (C) As in (B), but with a 5′ DNA (left) or RNA (right) overhang from two independent experiments. Insets indicate half-life quantification. (D) Representative curves from FAD with four different 3′ overhang substrates from two independent experiments. For clarity, only + ATP curves are shown. Inset indicates half-life quantification.

Since cellular RNAs are structured, we next monitored UPF1-HD dissociation from structured substrates. Although the greater mass of hairpin substrates limited the dynamic range of the assay, we observed ATP-stimulated dissociation (Supplementary Figure S5D). As with substrates containing 3′ ssDNA overhangs, UPF1-HD dissociation rates from hairpin substrates plateaued at longer substrate lengths (Supplementary Figure S5E). As an orthogonal approach, we developed a simultaneous unwinding and dissociation (SUD) assay by modifying the unwinding assay to include fluorescent UPF1-HD (UPF1-HD-647) and a second quencher strand to capture and reduce the fluorescence of UPF1-HD-647 after dissociation from the substrate (Supplementary Figure S5F). This assay recapitulated ATP-stimulated dissociation of UPF1-HD, with comparable kinetics (Supplementary Figure S5G). Consistent with a processivity < 200 nt on hpDNA and the results of the FAD assay (Supplementary Figure S5E), UPF1-HD dissociation rates plateaued at longer hairpin lengths (Supplementary Figure S5H). Taken together, these assays allow direct measurement of UPF1-HD dissociation from nucleic acids and indirectly confirm processivity <200 nt.

UPF1 ATPase activity is tightly coupled to unwinding and translocation, and dissociation is driven primarily by ATP hydrolysis

To begin to dissect relationships among UPF1 RNA binding, ATPase activity, and mechanochemical coupling, we first performed NADH-coupled ATPase assays (29) (Supplementary Figure S6A–C). In conjunction with the fluorescence-based unwinding assays described above, ATPase measurements allow quantification of the ability of UPF1 to couple ATP hydrolysis with work in the form of unwinding and translocation. We observed tight coupling of ∼1 ATP hydrolyzed per nucleotide unwound or translocated (Supplementary Figure S6D). As controls, we used DE636AA and K498A mutants, both of which expectedly lacked detectable ATPase activity (Supplementary Figure S6E, F). We next quantified nucleic acid binding affinity using fluorescence anisotropy. In these equilibrium binding assays, we obtained an apparent K_D of 0.5–2 nM for UPF1-HD in the absence of ATP (Supplementary Figure S6G, H), representing a binding affinity 1–2 orders of magnitude tighter than previously reported fluorescence anisotropy binding assays (38,47). To investigate this discrepancy, we considered known factors that limit apparent equilibrium binding affinities, including substrate concentrations that exceed the K_D and incubation times that are insufficient to reach equilibrium (57). Indeed, we found that letting the system reach equilibrium by increasing incubation time (Supplementary Figure S6I, J) and decreasing substrate concentration (Supplementary Figure S6K-L), both significantly increased the measured binding affinity of UPF1-HD for nucleic acids.

The relative importance of ATP binding and ATP hydrolysis for regulation of UPF1-HD association and dissociation from RNA are currently unclear. In agreement with previous equilibrium binding assays (21,38,43,44), we found that ATP binding (using the non-hydrolyzable ATP analog, AMP-PNP) reduced UPF1-HD binding affinity for nucleic acids, but not to the same extent as ATP hydrolysis (using ATP, Supplementary Figure S7A–C). Furthermore, consistent with previous kinetics experiments (16), only ATP, not AMP-PNP (or ATPγS or ADP), promoted dissociation of UPF1-HD from substrates in the FAD assay (Supplementary Figure S7D–F). These data indicate that ATP binding may impair UPF1 association with RNA but that ATP hydrolysis is the primary driver of UPF1 dissociation from RNA.

In silico Rosetta-Vienna RNP ΔΔG method identifies known and novel mutations affecting UPF1 RNA binding properties

With a quantitative view of UPF1-HD enzymatic activities in vitro, we next set out to discover new UPF1-HD mutants with alterations in enzymatic activities and characterize them alongside previously identified ATPase and helicase mutants using the high-throughput assays we developed. Therefore, we used the Rosetta-Vienna RNP ΔΔG method (39,40) to computationally identify UPF1-HD point mutants with RNA binding energies significantly different from WT UPF1-HD (Figure 4A). The predicted RNA binding energies matched expectations for previously characterized ‘grip mutants’ found to have significantly lower processivities in magnetic tweezers experiments (A546H/K547P/S548A [AKS-HPA], A546H, K547P, and R549S) (46), indicating the validity of this approach (Figure 4B). Of the 11 818 point mutants tested, ∼96% had no effect on predicted RNA binding energy (−1 < ΔΔG < 1, in units of kcal/mol), and most predicted large effects trended towards weaker binding (ΔΔG > 1), with only 29 mutations with predicted stronger binding (ΔΔG < −1, Supplementary Figure S8A, Table S3). We characterized several of these mutants in vitro, then selected a subset for cellular characterization, including two novel mutants with −ΔΔG predictions (E797R, G619K), four of the previously characterized grip mutants (46) (all +ΔΔG predictions), and the R843C dominant-negative mutant (16,58) (Figure 4C).

Figure 4.

Rosetta-Vienna ΔΔG method identifies previously characterized grip mutants and novel UPF1-HD mutants with altered RNA binding properties. (A) Workflow of in silico screen to identify UPF1 mutants of interest (see Materials and Methods). (B) Change in the binding free energy compared to WT (ΔΔG) predicted by the screen for previously identified mutants. ΔΔG for each mutant was obtained through comparison against the 20 lowest energy relaxed WT structures. (C) Structure of UPF1-HD indicating selected mutants. Colors correspond to classification in (D). (D) Qualitative Venn diagram generated with DeepVenn (62) of mutants from in vitro characterization (see Supplementary Table S3 for quantitative values).

A546H and G619K/A546H UPF1-HD mutants exhibit severe biochemical defects

After purifying (Supplementary Figure S8B, C) and characterizing these mutants, we identified several overlapping enzymatic classes, including UPF1-HD mutants with altered ATPase rates, processivities, unwinding rates, and ATPase coupling efficiencies (Figure 4D). Aside from the ATPase-deficient mutants, the most severe defects were observed with the previously characterized grip mutant, A546H (AKS-HPA showed similar defects). Consistent with previous reports (46), A546H exhibited significantly lower processivity and slower unwinding rate (Supplementary Figure S8D, top; Supplementary Table S4). A546H also showed ∼25-fold less efficient coupling between ATP hydrolysis and unwinding and modestly lower equilibrium binding affinity (Supplementary Figure S8E, top; Supplementary Table S4). Additionally, A546H showed faster ATP-independent dissociation as monitored by the FAD assay but continued to undergo ATP-stimulated dissociation from 3′ ssDNA overhang substrates (Supplementary Figure S8F, top; Supplementary Table S4). In contrast, A546H did not undergo ATP-stimulated dissociation from substrates with 5′ ssDNA overhangs, even though A546H ATPase activity was enhanced when bound to such structured substrates (Supplementary Figure S8G, top; Supplementary Table S4). This effect was even more pronounced on substrates with 5′ ssRNA overhangs (Figure 5A), in which ATP simulated dissociation of WT UPF1-HD (Figure 5B, top), but slowed down dissociation of A546H from these substrates (Figure 5B, middle). We reasoned that this could be caused by decreased dissociation of A546H molecules that had translocated into the duplex region. The A546H half-life on the substrate would thus consist of two populations, one undergoing fast dissociation from single-stranded regions and one undergoing slow dissociation from the duplex region. In the absence of ATP, A546H would be unable to access the duplex region, so fast dissociation from the ssRNA would predominate.

Figure 5.

Dissociation and PIFE assays indicate differential release kinetics of UPF1-HD mutant proteins. (A) FAD assay with indicated 5′ RNA overhang substrate. (B) Representative curves from FAD of WT (top, same as right panel of Figure 3C), A546H (middle), and GA-KH (bottom) UPF1-HD constructs from 2 independent experiments. (C) Schematic of PIFE assay with indicated 5′ RNA overhang substrate. (D) Representative curves from PIFE assay of WT (top), A546H (middle), and GA-KH (bottom) from 2 independent experiments. Trap control reactions were performed by adding 1000-fold excess trap strand prior to UPF1-HD addition. (E) Fluorescence-based biotin-streptavidin displacement assay. (F) Representative curves of biotin-streptavidin displacement assay of WT (top), A546H (middle) and GA-KH (bottom) from 2 independent experiments. Quencher control reactions were performed by adding 63-fold excess biotin-quencher strand prior to streptavidin addition. (G, H) Half-life quantification from (B) and (D). (I) Fraction streptavidin displaced quantified from (F) and ATPase coupling calculated by dividing unwinding rate by ATPase rate on hpDNA. Error bars indicate standard deviation.

To test this hypothesis, we monitored the presence of UPF1-HD near the single-strand/double-strand junction using protein-induced fluorescence enhancement (PIFE), a phenomenon caused by enhanced fluorescence intensity of dyes such as Cy3 that undergo cis-trans isomerization when in proximity to proteins (Figure 5C) (59). In control PIFE assays, WT UPF1-HD bound at the junction caused enhanced Cy3 fluorescence intensity, which decreased over time only in the presence of ATP as the protein dissociated from or translocated through the RNA (Figure 5D, top). Consistent with impaired ATP-stimulated dissociation from the duplex region, A546H did not exhibit ATP-stimulated PIFE decrease over time (Figure 5D, middle). Together, the ATPase, FAD, and PIFE assays suggest A546H cannot efficiently release from structured regions, where it instead unproductively hydrolyzes ATP.

Since A546H did not efficiently couple ATP hydrolysis into work in the form of unwinding, we reasoned that this should also result in reduced force generation and, in turn, impaired displacement of proteins from nucleic acids. As a stringent test of protein displacement, we developed a fluorescence-based biotin-streptavidin displacement assay, in which fluorescence decreases as fluorescent streptavidin is displaced from the biotinylated DNA (Figure 5E). Consistent with previous reports (28,29), WT UPF1-HD efficiently disrupted biotin-streptavidin interactions only in the presence of ATP (Figure 5F, top). Importantly, A546H streptavidin displacement activity was significantly less efficient than that of WT UPF1-HD (Figure 5F, middle), consistent with the idea that it generates less work from each ATPase cycle.

In an attempt to enhance the helicase activity of the A546H mutant, we combined it with the G619K mutation, which exhibited increased rates of ATPase, unwinding, and dissociation, to generate G619K/A546H (GA-KH). Relative to A546H, GA-KH showed modestly enhanced unwinding activity, slightly faster unwinding rate, and similar decoupling of ATPase activity from unwinding (Supplementary Figure S8D-E, bottom; Supplementary Table S4). Notably, GA-KH exhibited enhanced ATP-stimulated dissociation relative to A546H, as shown by both FAD (Figure 5B, bottom; Figure 5G, Supplementary Figure S8F, G, bottom) and PIFE (Figure 5D, bottom; Figure 5H) assays. Consistent with decoupled ATPase activity from unwinding, GA-KH exhibited similar streptavidin displacement activity compared to A546H (Figure 5F, bottom; Figure 5I). Taken together, both A546H and GA-KH exhibit similar biochemical defects, although GA-KH has modestly higher unwinding activity and restores ATP-simulated dissociation.

Slow and poorly processive UPF1 mutants with mechanochemical decoupling support efficient NMD in human cells

After quantifying the biochemical aspects of A546H and GA-KH in addition to other UPF1-HD mutants (Figure 4D, Supplementary Figure S9A–D, Table S4), we set out to determine their NMD efficiency in human cells. To this end, we generated stable cell lines constitutively expressing siRNA-resistant CLIP-tagged UPF1 (CLIP-UPF1) mutants in Flp-In T-Rex 293 cells, with or without siRNA-induced endogenous UPF1 knockdown (Figure 6A-B, Supplementary Figure S10A–D). We used RT-qPCR to detect mRNA levels of EJC-dependent (SRSF2, 3 and 6 transcript isoforms containing PTCs) and 3′UTR EJC-independent (SMG1, 5, 6, 7 and UPF2 transcripts) NMD targets. We verified that UPF1 knockdown and rescue with WT CLIP-UPF1 resulted in marked upregulation of NMD-sensitive mRNAs and subsequent restoration back to control levels, respectively (Figure 6C, D, compare siNT CLIP only to siUPF1 CLIP only to siUPF1 WT).

Figure 6.

RT-qPCR reveals that UPF1 mutants with impaired helicase activity support NMD in human cells. (A) Workflow of knockdown/rescue experiments using Flp-In T-Rex 293 stably expressing CLIP-UPF1 mutants. Dox was added to induce expression of CLIP-UPF1 mutants in (E). (B) Western blot of endogenous UPF1 and CLIP-UPF1 expressed at endogenous levels upon siUPF1 treatment from 2 independent experiments. β-actin (bottom blot) was used as a loading control. White space after WT lane indicates removal of irrelevant lanes of the same membrane, and black outlines indicate separate membranes. Quantification of relative CLIP-UPF1 and endogenous UPF1 is shown in Supplementary Figure S10B. (C, D) EJC-dependent (C) and long 3′ UTR-dependent (D) NMD target mRNA levels relative to GAPDH using RT-qPCR normalized to the siNT CLIP condition. siNT is non-targeting siRNA, and siUPF1 is UPF1 siRNA. CLIP-UPF1 expression was similar to endogenous UPF1 expression as shown in Supplementary Figure S10A. Error bars indicate standard deviation from 3 independent experiments. (E) As in (C, D) but with CLIP-UPF1 mutants expressed ∼6-fold above endogenous UPF1 levels and normalized to the siNT GFP condition, from three independent experiments. Statistical significance was determined by two-way ANOVA, with comparison between the indicated conditions and siNT with CLIP only or GFP rescue (*P < 0.05, Supplementary Table S5).

We next assayed several of the characterized mutants using this approach. As a control based on previous work, R843C was deficient in restoring NMD (16,58), although efficiencies varied by transcript (Figure 6C, D). Despite the severe biochemical defects of the two grip mutants (AKS-HPA, A546H), they were partially NMD functional (Figure 6C, D). Mutants that were slower (K547P, E797R, GA-KH), faster (G619K), less processive (E797R, GA-KH), or with decoupled ATPase activity (K547P, R549S, G619K, GA-KH) restored NMD to WT or near WT levels (Figure 6C-D). Importantly, GA-KH exhibited similar biochemical defects to A546H, with the exception of modestly higher unwinding activity and restored ATP-stimulated dissociation from structured RNA. This dataset, in particular the ability of GA-KH to support NMD, provides strong evidence that UPF1 does not need to processively remodel mRNPs to promote NMD initiation in human cells.

Finally, we determined whether overexpressing UPF1 mutants (Supplementary Figure S10A, C, E) would enhance NMD efficiency. Consistent with previous reports (42,60), we found that overexpression of WT was not sufficient to further down-regulate canonical NMD target mRNAs (Supplementary Figure S10F, G, compare siNT CLIP only or GFP to siNT WT). Furthermore, mutants with impaired NMD activity under near-endogenous expression conditions (R843C, AKS-HPA, A546H) did not lead to enhanced rescue (Figure 6E). However, these mutants led to significant upregulation of endogenous UPF1 mRNA levels in the siNT conditions (Supplementary Figure S10E), suggesting that an excess of faulty UPF1 causes feedback upregulation of endogenous UPF1 (42,60). Together, this quantitative characterization (Figure 7A, Supplementary Table S4) points toward a model in which UPF1 processivity, unwinding rate, and mechanochemical coupling can be significantly impaired without disrupting regulation of well-characterized NMD targets (Figure 7B). These findings are consistent with models in which UPF1 uses ATP hydrolysis to sample potential substrates and perform local mRNP remodeling events to promote NMD.

Figure 7.

In vitro and cellular characterization reveal UPF1 mutants with slow unwinding rate, poor processivity, and poor ATPase coupling, support NMD in human cells. (A) Heatmap of mutant properties quantified from >2 independent binding assays (nucleic acid binding affinity columns), single and multiple turnover unwinding assays (processivity columns; values from best-fit curves), 10 bp hpDNA single turnover unwinding assays (unwinding rate column), ATPase assays (ATPs/sec columns), FAD assays (residence time and ATP-stimulated dissociation columns), streptavidin displacement, and relative NMD rescue from cellular experiments (see Materials and Methods). ATPs/bp is calculated by dividing ATPs/s on hpDNA by unwinding rate. ATP-stimulated dissociation is calculated by dividing -ATP half-life by + ATP half-life on 5′ or 3′ DNA overhang substrates. Colors are relative to each column. More purple indicates greater activity detriments. VG-ED is V689E/G692D, EA-RH is E797R/A546H, GA-KH is G619K/A546H, EK-RP is E797R/K547P. See Supplementary Table S4 for more details. (B) UPF1 enzymatic activities sufficient for efficient mammalian NMD. ATP-dependent UPF1 activities assayed in this work are shown, with properties of WT and GA-KH UPF1-HD indicated, both of which support full NMD activity on well-characterized target mRNAs.

DISCUSSION

UPF1 ATPase activity has long been recognized as critical for NMD. However, the precise mechanisms by which UPF1 biochemical properties derived from ATPase activity influence NMD remain uncertain. The quantitative high-throughput assays developed here enable the dissection of UPF1 properties including RNA binding, ATP hydrolysis, translocation, unwinding, protein displacement, and ATPase-stimulated dissociation from RNA (Figure 7). Our results reconcile reports of high UPF1 processivity (Supplementary Figure S3) (28,46–48) with longstanding observations that UPF1 undergoes ATP-stimulated dissociation from nucleic acids (Supplementary Figure S7) (17,18,20,21,25,27,38,47). We find that UPF1 binds very tightly to nucleic acids in the absence of ATP (∼0.5 nM K_D, Supplementary Figures S6 and S7) but that ATP hydrolysis strongly promotes UPF1 dissociation (Supplementary Figure S7). This behavior results in moderate processivity (< 200 bp) when duplex-destabilizing forces are not applied to the substrate (Figures 1 and 2).

Helicase-deficient UPF1 mutants support efficient mammalian NMD

By combining our battery of biochemical assays with cellular knockdown-rescue experiments to study a panel of known and novel UPF1 mutants, we were able to define biochemical requirements for UPF1 function on canonical NMD substrates (Figure 7). We identify UPF1 mutants with ATP hydrolysis rates similar to or higher than WT UPF1 but, because of reduced mechanochemical coupling, have impaired unwinding and protein displacement activities. In particular, AKS-HPA, A546H, and GA-KH mutants unwind 1 bp per ∼20–30 ATPs hydrolyzed, resulting in slow (∼0.1–0.25 bp/s) and poorly processive (∼10–20 nt) unwinding and ∼4-fold lower streptavidin displacement activity. These mutants nevertheless largely or fully rescue regulation of well-characterized NMD targets stabilized by knockdown of endogenous UPF1 (Figures 6 and 7). Our data suggest that the process of NMD initiation does not require UPF1 to be an effective helicase. However, they do not rule out the possibility that decay initiation on some substrates or complete turnover of decay intermediates require more robust UPF1 helicase activities.

Butterfly versus bulldozer models of UPF1 in NMD

Biochemical studies of UPF1 have led to two divergent models for potential roles of UPF1 enzymatic activity in NMD. In one view, which we term the butterfly model, UPF1 undergoes frequent rounds of binding and ATPase-stimulated dissociation as it surveys the transcriptome for NMD targets (35). If UPF1 spends enough time on an RNA to sense a translation termination event and scaffold assembly of decay complexes, decay may be initiated (14,23). This view is biochemically supported by the slow unwinding rate of human UPF1 (∼1 bp/s) in vitro (Supplementary Figure S3A) (28,46,47) and the longstanding observation that UPF1 tends to release RNA upon ATP binding/hydrolysis (Supplementary Figure S7) (17,18,20,21,25,27,38,47). In further support of this model, ATPase-deficient UPF1 mutants lose substrate selectivity, becoming stably bound at non-productive sites throughout the transcriptome (16,35), implying the importance of ATPase-stimulated dissociation for substrate selection.

In the alternative view, which we term the bulldozer model, UPF1 uses ATP to processively remodel mRNPs for efficient NMD. Support for this model is derived from the ability of UPF1 to disrupt biotin-streptavidin interactions (Figure 5) (28,29), displace nucleic acid binding proteins (28), and from single-molecule magnetic tweezers experiments that revealed UPF1 can processively travel over 10,000 nt before dissociating (Supplementary Figure S3B, C) (28,46–48). A characterized ‘grip mutant’ (AKS-HPA) with severely impaired processivity in magnetic tweezers assays could not carry out NMD on the DAL7 transcript in S. cerevisiae, suggesting high UPF1 processivity may be important for efficient NMD (46). However, because nucleic acid substrates are under duplex-destabilizing forces in magnetic tweezers experiments, it is unclear if high processivity is retained under physiological conditions. Notably, force-dependent processivity enhancement (61) has been reported for the UPF1-like SARS-Cov-2 Nsp13 RNA helicase (52) along with other SF1 (49–51), SF2 (54) and SF4 (53) helicases.

As in previous reports that used magnetic tweezers (28,46–48), we observed very high UPF1 processivity in this context (Supplementary Figure S3B, C). However, UPF1 exhibited processivity of only 10s-100s of nt in ensemble and SPRNT assays performed without duplex-destabilizing forces applied to the nucleic acid substrate (Figures 1 and 2, Supplementary Figures S1–S4). Correspondingly, ATPase-stimulated dissociation kinetics of UPF1 from nucleic acids using FAD, SUD, and PIFE assays were consistent with moderate processivity (Figures 3 and 5, Supplementary Figure S5). UPF1 appears to be a striking example of force-dependent processivity enhancement, as measured processivity differs by 2–3 orders of magnitude in the presence and absence of duplex-destabilizing forces (Figures 1 and 2, Supplementary Figures S1–S4) (28,46–48). These forces may lower the energy barrier to forward translocation of UPF1, dramatically increasing processivity (61). If this energy barrier corresponds to a short-lived state in the ATPase cycle, unwinding rate would remain largely unchanged, as we observed (Supplementary Figure S3A).

Mechanochemically decoupled UPF1 mutants support the butterfly model

Overall, our data support the butterfly model, which requires a balance between stable UPF1 engagement with target RNAs and efficient release from non-target RNAs (35). UPF1 can achieve this balance through ATPase-mediated control of RNA binding, as UPF1 is capable of binding RNA with sub-nanomolar affinity (Supplementary Figures S6 and S7) but can readily release RNA upon ATP hydrolysis. Without ATPase activity driving dissociation from RNA, UPF1 cannot perform proper substrate selection, instead becoming non-productively locked on mRNAs across the transcriptome (16,35). Importantly, the robust NMD activity of mechanochemically decoupled UPF1 mutants differs markedly from that of NMD-deficient mutants with low (R843C) or no ATPase activity (previously studied ATPase-dead DE636AA, K498A and others) (14,15,17,20–23).

We hypothesize that a key difference between ATPase-deficient UPF1 mutants and mechanochemically decoupled mutants is the ability to undergo ATPase-stimulated dissociation. AKS-HPA, A546H and GA-KH mutants underwent ATPase-stimulated dissociation from unstructured regions (Supplementary Figure S8 and S9). However, differential behavior on structured regions may explain the greater NMD activity of GA-KH, as A546H inefficiently dissociated from structured regions, whereas ATPase-stimulated dissociation from structured regions was partially restored in the GA-KH double mutant (Figure 5, Supplementary Figure S8G). While we cannot rule out the contributions of other slight biochemical differences between A546H and GA-KH (Supplementary Figure S8), the increased NMD activity of GA-KH is consistent with a role for ATPase-stimulated dissociation in maintaining proper NMD.

UPF1 ATPase-mediated release from RNA as a control point for NMD

Increasing evidence suggests that mechanisms for manipulation of ATPase-stimulated dissociation of UPF1 are important for cellular control of NMD specificity. For example, the UPF1 regulatory loop that protrudes into the RNA-binding channel mediates sensitivity to protective RNA binding proteins such as PTBP1 (29,42,47). This may be due PTBP1 exploiting certain conformational states of UPF1 to promote dissociation. Conversely, activators of UPF1-mediated mRNA decay such as Staufen and histone stem loop binding protein may promote stable association of UPF1 with mRNPs by suppressing ATPase-stimulated dissociation from target mRNAs (3). The experimental framework developed here will enable future investigation to elucidate the biochemical mechanisms of trans-acting factors on UPF1.

DATA AVAILABILITY

All data are provided in the included main text and supplementary figures and tables.

SUPPLEMENTARY DATA

Supplementary Data are available at NAR Online.

FUNDING

J.M.C. and J.H.G. were supported by National Human Genome Research Institute Grant [R01HG005115]; J.H.C., C.D.W., K.C.N. and J.R.H. were supported by the Intramural Research Program, National Heart, Lung, and Blood Institute, National Institutes of Health. Funding for open access charge: National Heart, Lung, and Blood Institute.

Conflict of interest statement. J.M.C., J.H.G. and the University of Washington hold a patent on the SPRNT technology (US patent no. 10359395).