Abstract
RNA stability is meticulously controlled. Here, we sought to determine whether an essential post-transcriptional regulatory mechanism plays a role in pain. Nonsense-mediated decay (NMD) safeguards against translation of mRNAs that harbor premature termination codons and controls the stability of ∼10% of typical protein-coding mRNAs. It hinges on the activity of the conserved kinase SMG1. Both SMG1 and its target, UPF1, are expressed in murine DRG sensory neurons. SMG1 protein is present in both the DRG and sciatic nerve. Using high-throughput sequencing, we examined changes in mRNA abundance following inhibition of SMG1. We confirmed multiple NMD stability targets in sensory neurons, including ATF4. ATF4 is preferentially translated during the integrated stress response (ISR). This led us to ask whether suspension of NMD induces the ISR. Inhibition of NMD increased eIF2-α phosphorylation and reduced the abundance of the eIF2-α phosphatase constitutive repressor of eIF2-α phosphorylation. Finally, we examined the effects of SMG1 inhibition on pain-associated behaviors. Peripheral inhibition of SMG1 results in mechanical hypersensitivity in males and females that persists for several days and priming to a subthreshold dose of PGE2. Priming was fully rescued by a small-molecule inhibitor of the ISR. Collectively, our results indicate that suspension of NMD promotes pain through stimulation of the ISR.
SIGNIFICANCE STATEMENT Nociceptors undergo long-lived changes in their plasticity which may contribute to chronic pain. Translational regulation has emerged as a dominant mechanism in pain. Here, we investigate the role of a major pathway of RNA surveillance called nonsense-mediated decay (NMD). Modulation of NMD is potentially beneficial for a broad array of diseases caused by frameshift or nonsense mutations. Our results suggest that inhibition of the rate-limiting step of NMD drives behaviors associated with pain through activation of the ISR. This work reveals complex interconnectivity between RNA stability and translational regulation and suggests an important consideration in harnessing the salubrious benefits of NMD disruption.
Introduction
mRNA control permeates neuronal plasticity (H. Wang and Tiedge, 2004; Costa-Mattioli et al., 2009; de la Peña et al., 2019; Loerch et al., 2021). Nonsense-mediated decay (NMD) is a prototypical mechanism of mRNA regulation (Kurosaki and Maquat, 2016). It was discovered as a surveillance pathway that triggers degradation of mRNAs that harbor premature termination codons (PTCs) (Losson and Lacroute, 1979; Maquat et al., 1981). mRNAs that harbor PTCs are the result of multiple processes including mutation and splicing defects (Wei-Lin Popp and Maquat, 2013). Translation of aberrant transcripts can result in the production of toxic polypeptides (Kugler et al., 1995). Multiple mechanisms have been proposed to account for the specificity of NMD. In the exon junction complex (EJC) model, the EJC is deposited at exon junctions joined together because of splicing. Stop codons are generally found in the last exon and are therefore not followed by an EJC (Nagy and Maquat, 1998). If a ribosome encounters a stop codon before an EJC, this abnormal arrangement serves as a signal for NMD (Cheng et al., 1994; Carter et al., 1996; Thermann et al., 1998; Zhang et al., 1998). In contrast, the long 3′ untranslated region (UTR) model posits that a critical NMD factor designated upstream frameshift 1 (UPF1) interacts with the 3′ UTR of mRNA in a length-dependent manner (Amrani et al., 2004; Bühler et al., 2006; Hogg and Goff, 2010; Shigeoka et al., 2012). As a result, mRNAs with long 3′ UTRs are preferentially degraded by NMD. However, specific RNA-binding proteins can counteract NMD (Ge et al., 2016; Kishor et al., 2019). A second EJC-independent mechanism involves the observation that highly structured mRNAs tend to have shorter half-lives in mammals (Sun et al., 2019). Disruption of an RNA-binding protein that recognizes structured elements, G3BP Stress Granule Assembly factor 1 (G3BP1), or UPF1 stabilizes mRNAs with highly structured 3′ UTRs (Fischer et al., 2020). A major goal of this work is understanding which mRNAs are controlled by NMD specifically in sensory neurons.
Mechanistic insights into NMD have enabled the generation of numerous small-molecule inhibitors. NMD is initiated by the association of a core complex containing UPF1-3 (Gehring et al., 2005). When the complex is assembled on RNA, SMG1 phosphorylates the core member of the complex UPF1 (Yamashita et al., 2001; Conti and Izaurralde, 2005; Lejeune and Maquat, 2005). This is the rate-limiting step of NMD; and as a result, SMG1 inhibitors have been used to inhibit NMD. (Gopalsamy et al., 2012). Following UPF1 phosphorylation, enzymes that stimulate mRNA metabolism and decay are recruited (Lejeune et al., 2003). Thus, SMG1 activity is critical to NMD.
Another major pathway of translational control is the integrated stress response (ISR). Numerous stimuli, including double-stranded RNA, proteotoxic stress, and reactive glycolytic metabolites, all activate the ISR (Pakos-Zebrucka et al., 2016; Barragán-Iglesias et al., 2019). The kinases that mediate the ISR converge on phosphorylation of eukaryotic initiation factor 2α (eIF2α) (Donnelly et al., 2013). eIF2, a trimeric protein complex, is tasked with delivery of the initiator methionine tRNA to the 40S subunit when bound to GTP. eIF2α is the regulatory subunit of the complex (Schmitt et al., 2012). When it is phosphorylated, eIF2α prevents eIF2B from promoting guanine exchange on free eIF2 and results in translational repression of most mRNAs (Gordiyenko et al., 2019). Specific mRNAs evade repression by eIF2α phosphorylation, most notably the activating transcription factor 4 (ATF4), which encodes a master regulator of the ISR (Ameri and Harris, 2008; Magne et al., 2011). Dephosphorylation of eIF2α is controlled by constitutive repressor of eIF2-α phosphorylation (CReP) phosphatase (Harding et al., 2009).
Here, we examined the role of NMD in pain signaling. Pharmacologic inhibition of NMD induced mechanical hypersensitivity concomitant with induction of eIF2α phosphorylation. The nociceptive consequences of NMD inhibition were alleviated by inhibition of the ISR. Collectively, the data reveal that disruption of NMD causes mechanical hypersensitivity through induction of the ISR.
Materials and Methods
Animal model
WT male Swiss Webster mice were obtained from Taconic Laboratories for both in vitro primary DRG cultures (4-6 weeks old) and in vivo behavioral testing (8-12 weeks old). Animals were housed in a humidity- and climate-controlled vivarium with 12 h light/dark cycles and unlimited access to food and water. All animal procedures were authorized by the Institutional Animal Care and Use Committee at the University of Texas at Dallas and the University of Wisconsin-Madison and followed the guidelines set by the International Association for the Study of Pain.
Primary DRG cultures
DRGs were extracted from 4- to 6-week-old mice and stored in cooled HBSS (Fisher Scientific, catalog #14170112). Following dissection, DRGs were enzymatically dissociated using collagenase A (1.0 mg/ml, Millipore Sigma, catalog #10103578001) at an incubation temperature of 37°C for 25 min. This was followed by treatment with collagenase D (1.0 mg/ml, Millipore Sigma, catalog #11088858001) and papain (30 U/ml, Millipore Sigma, catalog #10108014001) at 37°C for 20 min. DRGs were gently triturated in a 1:1 combination of 1 mg/ml trypsin inhibitor (Millipore Sigma, catalog #10109878001) and BSA (Fisher Scientific, catalog #J10856.22), then filtered through a 70 µm cell strainer (Corning, catalog #431751). Cells were gently pelleted (400 × g for 4 min) and resuspended in DRG culture media, which consists of the following components: DMEM/F12 with GlutaMAX (Fisher Scientific, catalog #10565018), 10% FBS (Fisher Scientific, catalog #26140079), 1% penicillin and streptomycin, 5 ng/ml NGF (R&D Systems, catalog #1156-NG-100), and 3 µg/ml 5-fluorouridine with 7 µg/ml uridine. NGF was supplemented to promote neuronal survival, while 5-fluorouridine and uridine were used to suppress mitosis and restrict proliferation of support cells. Cultures were incubated at 37°C in a humidified chamber maintained with 5% CO2. Media was exchanged on alternating days.
Library generation and sequencing
Each sample corresponded to DRGs isolated from cervical (C1) to lumbar (L5) spinal segments from 10 mice. Primary cultures were plated in 100-mm-diameter poly-D-lysine-coated culture dishes (BD Falcon, catalog #353003) and stabilized for 5 d. On day 6, cells were treated with the SMG1 inhibitor, 11j (1 μm for 4 h; Chemieliva Pharmaceutical) (Gopalsamy et al., 2012). Total RNA was isolated using TRIzol reagent (Fisher Scientific, catalog #15596026) and applied to a Phase Lock Gel-Heavy column (PLG, QuantaBio, catalog #2302830). Eluted RNA was then rRNA depleted using the RiboCop rRNA kit, according to the manufacturer's protocols. Sequencing libraries were generated from 1 µg of RNA using Lexogen's QuantSeq 3′mRNA-Seq library kit. Library amplification was conducted based on endpoint qPCR using the PCR Add-on kit for Illumina (Lexogen, catalog #020.96). The final cycle number was chosen when the fluorescence value for each sample reached 50% of its maximum. The concentrations of purified libraries were quantified using Qubit (Invitrogen, catalog #Q10212), and the average size was determined by a fragment analyzer with a high-sensitivity NGS fragment analysis kit (Advanced Analytical Technologies, catalog #DNF-474-0500). Libraries were sequenced on an Illumina NextSeq500 Sequencer using 75 bp single-end high-output reagents (Illumina, catalog #20024906).
RNA-seq read mapping, quantification, and differential expression analysis
FastQ files were generated via the Illumina platform, downloaded to a Linux machine, and assessed for quality with FastQC 0.11.5 (Babraham Bioinformatics). Adaptor sequences were trimmed. Reads were mapped using TopHat 2.1.1 (with Bowtie 1.0.0) and the mouse reference genome (NCBI GRCm38.p4, Gencode vM10) (Langmead et al., 2009; Trapnell et al., 2009; NCBI Resource Coordinators, 2018; Frankish et al., 2019). After accounting for strand specificity, abundance was quantified with Cufflinks 2.2.1 with option-no-length-correction to prevent count length normalization (Trapnell et al., 2010). Abundance was measured in fragments per kilobase of transcript per million mapped reads and converted to transcript per million, for accurate comparisons between samples. Using TPM values, we were able to determine fold change (FC, defined as average abundance in treatment group divided by average abundance in vehicle group) and perform statistical analysis. False discovery rate (FDR) adjusted p values were calculated based on the Benjamini–Hochberg method to correct for multiple-hypothesis testing. An adjusted p value < 0.05 was considered significant. Upregulated transcripts were indicated by an FC > 1.5, while an FC < 0.66 was considered downregulated. The number of mapped reads per sample ranged from 20,690,113 to 47,447,515.
Informatics and publicly available datasets
Unless otherwise stated, all further informatic analyses/visualizations were performed in R with Rstudio (R Core Team, 2017; Rstudio Team, 2020). The R packages, biomaRt and party, were also used (Durinck et al., 2005, 2009; Strobl et al., 2007, 2008). The ViennaRNA package, specifically the RNAfold program, from Theoretical Biochemistry Group at the University of Vienna, was used to perform calculations regarding the secondary structure of RNA coding sequences and UTRs (Lorenz et al., 2011). Appropriate statistical measures were used in all comparisons. The details for statistical analyses can be found in corresponding Materials and Methods sections and figure legends.
Several publicly available datasets were used to define RNA characteristics. Axonal localization scores were obtained from a subcellular transcriptomic study of murine sensory neurons (Minis et al., 2014). Transcripts containing upstream open reading frames (uORFs) were identified by mining multiple datasets (Ingolia et al., 2011; Johnstone et al., 2016; Barragán-Iglesias et al., 2021). Data from single-cell sequencing were obtained from the Gene Expression Omnibus database with the key GSE59739 (Usoskin et al., 2015).
Single-cell analysis
Single-cell visualization was conducted based on published single-cell DRG sequencing data using R Seurat package 4.0 (Usoskin et al., 2015; Hao et al., 2021). Samples with empty wells were excluded from analysis in addition to any cells with >10% mitochondrial counts or fewer than 200 unique features. Normalization of the remaining cells was conducted using default Seurat parameters at a scale factor of 10,000. Unbiased clustering was done using the first 20 principal components at a resolution of 0.5, and after dimensional reduction resulted in six spatially distinct clusters. Cell type identity of each cluster was manually assigned based on expression of the following marker transcripts: Nefh (neurofilament, large-diameter neurons), Calca (peptidergic), Mrgprd (nonpeptidergic), Th (TH), and Vim (non-neuronal). Two spatially adjacent clusters identified as TH-expressing were combined into a single TH cell-type cluster resulting in five total clusters.
Immunohistochemistry (IHC)
Mice were anesthetized with isoflurane and killed by decapitation. Tissues (DRG, sciatic nerve, and spinal cord) were then harvested and flash frozen in OCT compound (Fisher Scientific, catalog #NC9806257) on dry ice. The tissues were sectioned at 20 µm and mounted onto SuperFrost Plus slides (Fisher Scientific, catalog #22-037-246). Slides were fixed with ice-cold 4% PFA (Fisher Scientific, catalog #28906) in TBS for 20 min, and subsequently washed with wash buffer (TBS + 0.025% Triton X-100), 3 times for 5 min each wash. Permeabilization was conducted using 0.3% Triton X-100 (Millipore Sigma, catalog #T8787-250ML) in TBS for 10 min followed by three consecutive washes with wash buffer, 5 min each wash. Sections were blocked for 2 h at room temperature with goat anti-mouse Fab Fragment (Jackson ImmunoResearch Laboratories, catalog #115-007-003) in TBS. The slides were washed once with wash buffer to remove unbound Fab fragments. The slides were then blocked with 10% NGS (Jackson ImmunoResearch Laboratories, catalog #005-000-121) and 1% BSA (Fisher Scientific, catalog #J10856.22) in TBS for 2 h. Primary antibodies were used to detect the following proteins: SMG1 (1:100, Santa Cruz Biotechnology, #SC-374557), peripherin (1:1000, Novus, #NBP1-05,423), NeuN (1:1000, Synaptic Systems, #266 004), CGRP (1:200, Enzo, BML-CA1134-0100), Isolectin B4 (1:1000, Invitrogen, #I21412), GFAP (1:1000, Invitrogen, #pa5-16291), and IBA1 (1:500, Synaptic Systems, #234009). Slides were incubated with primary antibodies at 4°C, overnight with gentle agitation. The slides were washed 3 times, and then appropriate secondary antibodies (AlexaFluor, Invitrogen) were applied for 1 h. Following additional washes, coverslips were mounted with ProLong Glass (Fisher Scientific, catalog #P36984). All images were acquired with an Olympus FV3000 Laser Scanning Confocal microscope at 20× magnification. Image analysis was done using ImageJ, and colocalization was analyzed using the JACop (Just Another Colocalization) function. Colocalization values are represented by Pearson's correlation coefficient (r).
Protein immunoblotting
For in vitro immunoblots, primary DRG neurons were cultured as described above. Cells from 5 mice were pooled and evenly distributed in a 6-well plate coated with poly-D lysine. On day 6, cells were treated with vehicle, 11j (1 μm), or 11j (1 μm) + ISR inhibitor (ISRIB) (200 nm). After the 4 h treatment, cells were washed with ice-cold PBS and then harvested in RIPA buffer (150 mm NaCl, 0.1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, 50 mm Tris-HCl, pH 8.0) supplemented with Pierce protease (Fisher Scientific, catalog #A32963) and phosphatase inhibitors (Fisher Scientific, catalog #A32957). Cell debris was isolated via centrifugation for 20 min at 18,000 × g at 4°C. Supernatants were then aspirated and used for immunoblotting. Protein concentration was determined via Pierce BCA assay (Fisher Scientific, catalog #23227). Samples were prepared in Laemmli sample buffer (Bio-Rad, catalog #1610747) and denatured following a 10 min incubation at 98°C. Samples were resolved on 10% SDS-PAGE gels and then transferred to methanol-activated Immobilon PVDF membranes (Millipore Sigma, catalog #IPVH00010). Membranes were blocked with 5% milk in TBS + 0.05% Tween-20 (TBST) for 2 h at room temperature, then incubated overnight at 4°C with primary antibodies diluted in blocking solution. Membranes were then washed 3 times with TBST followed by incubation with secondary antibodies diluted in blocking solution for 1 h at room temperature. Membranes were washed 3 times with TBST and then incubated with Immobilon ECL Ultra Western HRP Substrate (Millipore, catalog #WBULS0500) for 1 min before visualization using a ChemiDoc Touch Imaging System (Bio-Rad). Membranes were then stripped with Restore Plus Western blot stripping buffer (Fisher Scientific, catalog #46430) for 10 min and reblocked before probing with additional primary antibodies. This procedure was performed for the following primary antibodies: ATF4 (1:1000, Cell Signaling Technology, #11815S), p-eIF2α (1:1000, Cell Signaling Technology, catalog # 3398P), eIF2α (1:1000, Cell Signaling Technology, #9722), phosphorylated UPF1 (1:1000, Millipore, #07-1016), UPF1 (1:2000, Proteintech, #66898), and GAPDH (1:10,000, Cell Signaling Technology, #2118). Analysis was performed using Image Lab 6.0.1 (Bio-Rad). Phosphorylated proteins were normalized to their respective total proteins and expressed as a percent of change relative to vehicle groups.
Immunocytochemistry
Primary DRG neurons were cultured as described above. Cells from 2 mice were plated on 8-well chamber slides (LabTek) coated with poly-D-lysine. On day 6, cells were treated with vehicle (DMSO) or 11j (1 μm) for 4 h. The media was then aspirated, and cells were washed once with PBS before fixing in 4% formaldehyde for 15 min. Cells were then washed 3 times with PBS and permeabilized in 0.5% Triton X-100 for 10 min. Cells were again subjected to three PBS washes, then blocked in 10% NGS for 1 h. Primary antibodies against peripherin (1:1000, Novus, NBP1-05423) and CReP (1:300, St. John's Lab, STJ97266) diluted in 10% NGS were applied and incubated overnight at 4°C. Cells were washed 3 times with PBS. Goat anti-chicken-AF488 (1:1000, Invitrogen, A11039) and goat anti-rabbit-Cy5 (1:1000, Invitrogen, A10523) antibodies diluted in 10% NGS were applied to cells and incubated for 1 h. Cells were washed with PBS 3 times. Coverslips were mounted using ProLong Glass antifade mounting media (Fisher Scientific, catalog #P36984) and sealed with clear nail polish. Images were collected using a Leica SP8 point scanning confocal microscope using a 20× objective. Images were analyzed with ImageJ as follows. An ROI was manually drawn around each peripherin-positive cell body. Area, mean gray value, and integrated density were measured for each ROI. For each frame, five background ROIs outside of the peripherin-positive area were also measured. Corrected total cell fluorescence (CTCF) for each cell was calculated by subtracting the product of the area and the average mean gray value of the background ROIs from the integrated density, or CTCF = Integrated density – (area × average mean gray value of background ROIs).
qPCR analysis
To assay mRNA stability in a more direct manner, we used the transcriptional inhibitor actinomycin D (ActD) (Millipore Sigma, catalog #A4262) (Ratnadiwakara and Änkö, 2018). Cultured DRG neurons were treated with ActD (5 µg/ml) + vehicle or ActD + 11j (1 μm). After treatment, RNA was extracted and purified from cultured DRG neurons using GeneJET RNA Purification Kit (Fisher Scientific, catalog #K0702) following the manufacturer's guidelines. One microgram of RNA was reverse-transcribed to cDNA using ImProm-II reverse transcription system (Promega, catalog #A3800). The resulting cDNA was used for qPCR via iQ SYBR Green Supermix (Bio-Rad, catalog #1708880).
The primers used for qPCR are as follows: Atf4 (CCTAGGTCTCTTAGATGACTATCTGGAGG, CCAGGTCATCCATTCGAAACAGAGCATCG); Arc (AAGTGCCGAGCTGAGATGC, CGACCTGTGCAACCCTTTC); Ddit3 (CTGCCTGAACACGTCCACAT, CTGGAAGCCTGGTATGAGGAT); Gadd45b (CAACGCGGTTCAGAAGATGC, GGTCCACATTCATCAGTTTGGC); Srp72 (AGATATTCACACCCTAGCCCAA, AGAGACATGCTATCCGATGAGG); and Gapdh (TGGAGAAACCTGCCAAGTATGA, GGGATAGGGCCTCTCTTGCT). qPCR was conducted using the CFX96 Touch Real-Time PCR Detection System (Bio-Rad). Fold changes in expressions were quantified using the ΔΔCt method (where Ct is defined as the threshold cycle) relative quantification method (Livak and Schmittgen, 2001). Following quantification, values were normalized to Gapdh.
Behavioral assays
The behavioral effects of 11j were examined using the hyperalgesic priming model (Reichling and Levine, 2009). Mice received an intraplantar injection of 11j (1 or 6 µg) dissolved in 25 µl of physiological saline. Mice were then given 30 min to habituate to the apparatus. Mechanical sensitivity was measured using calibrated von Frey filaments (Ugo Basile, catalog #37450-275) and the up-down method (Dixon, 1980; Chaplan et al., 1994). von Frey filaments, with bending forces of 0.04, 0.07, 0.16, 0.4, 0.6, 1, 1.5, and 2 G, were applied to the mouse hind paw's plantar surface for 1-3 s. If the mouse lifts, shakes, or licks the hind paw while the filament is being applied, a paw withdrawal response is recorded. This was repeated for 5 times per mouse. Withdrawal threshold in grams was calculated by entering the results into a preprogrammed Excel sheet, which uses an algorithm that determines the force value that would elicit paw withdrawal in 50% of trials (10(Xf + kδ)/10,000, where Xf = value [in log units] of the final von Frey hair used, k = tabular value of positive and negative responses, and δ = mean difference [in log units] between stimuli). Various time points after the administration of 11j were measured (3 h, 6 h, 1 d, 3 d, 6 d). On day 9, after the animals had returned to their initial baseline thresholds, the development of priming was investigated by intraplantar administration of a subthreshold dose of PGE2 (100 ng; Cayman Chemicals, catalog #363-24-6). Mechanical sensitivity was then assessed 3 and 24 h after PGE2 injection.
To assess thermal sensitivity, we measured the paw withdrawal latency in response to a concentrated, radiant heat source using the Hargreaves test (Hargreaves et al., 1988). Mice were placed on a glass surface and allowed to habituate for 30 min. A focused high-intensity light beam was directed toward the mouse hind paw. The light intensity was set to 40% with a 20 s cutoff, according to the manufacturer's recommendations (IITC model 390). To reduce experimental variability, withdrawal latency measures were performed in triplicate with at least a 10 min gap between each trial. Thermal hypersensitivity was assessed 3 h after 11j and 3 h after PGE2 administration.
To examine the contributions of the integrated stress response on 11j-induced behavioral effects in mice, the ISRIB (1 or 10 µg) was intraplantarly coadministered with 11j. Mechanical and thermal sensitivity was then measured as described above. All treatments and measurements were randomized, and experienced experimenters who made the behavioral observations were blinded to the experimental groups.
Statistical analysis
In vitro data are presented as mean ± SEM and were gathered from at least three different cell cultures. Data from behavioral experiments are presented as mean ± SEM of at least 6 animals per group. Using G*power to calculate power with an 80% efficiency, expectations of a 50% effect size, and a 0.05 α setting, the sample size for behavior experiments was estimated to be n = 6. GraphPad Prism version 9.0 was used for plotting and statistical analysis (GraphPad Software). Unpaired t test was used to compare two independent groups. Statistical evaluation for three independent groups or more was performed by one-way or two-way ANOVA, followed by post hoc Dunnett or Tukey test, and the a priori level of significance at 95% confidence level was considered at p < 0.05. The significance of expression level changes before and after the treatment was calculated using a two-tailed Student's t test assuming unequal variances.
Data availability
All publicly available datasets referenced in the text have been outlined in Materials and Methods, along with any publicly available software. Code is available on reasonable request. High-throughput sequencing data have been deposited to Gene Expression Omnibus with accession code GSE211127 (token: mzqhskgkvtchbgp).
Results
NMD factors are present in the DRG
Phosphorylation of UPF1 by SMG1 is the rate-limiting step of NMD. Therefore, we examined the expression pattern of SMG1 and its target, UPF1, throughout the PNS. Our analysis of an existing single-cell dataset revealed five clusters corresponding to large-diameter neurons (Nefh, NF cluster), peptidergic neurons (PEP cluster), nonpeptidergic neurons (NP cluster), TH-expressing neurons (TH cluster), and non-neurons (Non-N cluster) (Fig. 1A) (Usoskin et al., 2015). When examining the expression of RNA in single-cell data, we based our analysis on two variables: average expression and percent expression. Average expression refers to the average abundance of the RNA across all the cells in a cluster. Percent expression refers to the number of cells in the cluster that the RNA is expressed in. Based on this classification, we found that SMG1 is in both neurons (32.15% expression) and non-neurons (12.03% expression). In neurons, SMG1 is expressed in neurofilament (51.45% expression), peptidergic (20.00% expression), nonpeptidergic (31.33% expression), and TH (25.83% expression) clusters. UPF1 was only detected in neuronal subtypes and was present in neurofilament (24.28% expression), peptidergic (15.56% expression), nonpeptidergic (19.28% expression), and TH (11.81% expression) clusters (Fig. 1B). The data suggest that SMG1 and UPF1 are transcribed in sensory neurons.
SMG1 and UPF1 are expressed in peripheral and spinal sites important for pain signaling. A, Single-cell clusters are based on the expression of markers for the following populations of cells: neurofilament present in large-diameter neurons (NF; Nefh), peptidergic neurons (PEP; Calca), nonpeptidergic neurons (NP; Mrgprd), TH (Th), and non-neuronal (Non-N; Vim). Data were obtained from Usoskin et al. (2015), which were then subjected to unbiased clustering. Right, Average expressions of SMG1 or UPF1 are indicated, from low (gray) to high (pink) levels of expression, in the different cell categories B, Quantification of coexpression of SMG1 or UPF1 with the different cell clusters. Average expression values have been normalized to better compare expression between clusters. SMG1 is expressed in both neuronal and non-neuronal cells, while UPF1 is only expressed in the neuronal subpopulation. C, IHC confirmed that SMG1 (magenta) is expressed in DRG neurons and colocalizes with CGRP (cyan), a marker for peptidergic neurons, IB4 (green) which labels nonpeptidergic neurons, and peripherin (yellow), a marker for small- and medium-diameter neurons which are mostly nociceptors. D, SMG1 is also present in the sciatic nerve and is colocalized with the markers CGRP, IB4, and peripherin. E, SMG1 is expressed in the dorsal horn of the spinal cord and colocalizes with the glial marker, GFAP, a microglial marker, IBA1, and the neuronal marker, NeuN. Colocalization is represented by Pearson's correlation coefficient (r).
To determine where SMG1 is translated, we performed IHC on L4-L5 DRG tissues. Following IHC, we calculated colocalization of SMG1 with our various sensory neuron markers using Pearson's correlation coefficient, which is denoted by r values (Fig. 1C). SMG1 is present in peptidergic neurons that express the CGRP (r = 0.49 ± 0.03), nonpeptidergic neurons defined by isolectin B4 (IB4) immunoreactivity (r = 0.22 ± 0.01), and in small-diameter peripherin-positive neurons (r = 0.62 ± 0.02) that demarcates Aδ and C fibers, which are mostly nociceptors (Fig. 1C) (Parysek and Goldman, 1988; Fornaro et al., 2008; Wainger et al., 2015). SMG1 is also present in the sciatic nerve and colocalizes with CGRP (r = 0.36 ± 0.11), IB4 (r = 0.41 ± 0.1), and peripherin (r = 0.31 ± 0.05) (Fig. 1D). It is expressed in the dorsal horn of the spinal cord and colocalizes with both GFAP-expressing glial cells (r = 0.15 ± 0.03) and immune cells that possess ionized calcium-binding adaptor molecule 1 (Iba1) (r = 0.29 ± 0.5). Additionally, SMG1 is present in spinal neurons that express neuron-specific nuclear (NeuN) protein (r = 0.41 ± 0.05) (Fig. 1E). These results indicate that SMG1 protein is present throughout the PNS, most notably in the soma and fibers of DRG sensory neurons. However, we were unable to identify specific tools under our assay conditions to quantify UPF1 expression throughout the PNS.
Pharmacologic disruption of SMG1 attenuates NMD
To probe the function of SMG1, we tested the effects of a small-molecule inhibitor on UPF1 phosphorylation and known NMD targets in sensory neurons. We made use of compound 11j, a highly selective and potent SMG1 inhibitor (Fig. 2A) (Gopalsamy et al., 2012). Treatment of primary DRG neurons with 11j reduced UPF1 phosphorylation by 90%, without affecting total UPF1 levels (Fig. 2B). We reasoned that, because UPF1 phosphorylation is required for NMD, inhibition of SMG1 should attenuate NMD activity. Accordingly, the stability of canonical NMD targets (i.e., Atf4, Arc, Ddit3, and Gad45b) was increased following the addition of 11j and a transcriptional inhibitor at both the 4 h (Fig. 2C) and 8 h time points (Fig. 2D) (Johnson et al., 2019). In contrast, the stability of a nontarget (Srp72) mRNA was insensitive to SMG1 inhibition (Fig. 2C,D) (Johnson et al., 2019). These results indicate that 11j triggers suppression of NMD in sensory neurons.
The SMG1 inhibitor, 11j, inhibits NMD in DRG neurons. A, A schematic representation of 11j's action on SMG1 kinase activity and its downstream phosphorylation of UPF1. B, Application of 11j to primary cultured DRG neurons decreased phosphorylated UPF1 (p-UPF1) levels, but not total UPF1 or GAPDH. Right, Quantification showing decreased p-UPF1/UPF1 levels in 11j-treated neurons; n = 6 cultures per treatment group. ****p < 0.0001 (two-tailed unpaired t test). C, D, qPCR quantification of mRNA levels reveals that 11j treatment stabilized the expression of canonical NMD targets (Atf4, Arc, Ddit3, Gadd45), 4 h (C) and 8 h (D) after transcriptional shutdown by ActD, whereas levels of the non-NMD target Srp72 were unchanged. *p < 0.05; **p < 0.01; ***p < 0.001; compared with relative vehicle-control group (multiple unpaired t test). Data are mean ± SEM, n.s. indicates that the comparison is not significant.
High-throughput sequencing reveals features of NMD targets in the DRG
To identify NMD targets globally in primary DRG sensory neurons, we analyzed RNA abundance in primary neurons treated with either a vehicle or 11j (Fig. 3A). After a 4 h exposure to 11j, the cells were harvested, and RNA was isolated. Ribosomal RNAs were depleted, and 3′ end sequencing was conducted using the QuantSeq 3′ mRNA-Seq protocol (Moll et al., 2014). Each group consisted of between three and four biological replicates. The correlation within groups was consistent (Pearson's R > 0.94). To identify transcripts whose abundance was affected by NMD inhibition, we calculated differential expression. We found that there were 787 upregulated (p < 0.05, FC > 1.5) transcripts and 442 downregulated (p < 0.05, FC < 0.66) transcripts (Fig. 3A). This corresponds to ∼9.1% of the transcripts we were able to detect. The result that there are more upregulated transcripts than downregulated is unsurprising given that NMD is a mechanism that reduces RNA stability. We suspect that the downregulated transcripts are the result of indirect effects. A possible explanation for this observation is that 32 RNA-binding proteins, and 46 transcription factors are increased following inhibition of NMD. Loss of activating factors could drive repression of many of these targets. We focus the bulk of our analyses on the upregulated group. Because our measurements are based on steady-state abundance and not RNA half-life, we are able to capture both direct and indirect targets of NMD. This is important as both may contribute to the biological functions of NMD in vivo.
Analysis of high-throughput sequencing reveals defining features of NMD targets in the DRG. A, Schematic represents the sequencing procedure and the resultant volcano plot indicating differential expression. Points represent individual transcripts. Axes represent adjusted p value (y) with a cutoff at –log10(0.05) and fold change (x) with cutoffs at log2(0.66 and 1.5). B, CDF of the log2(fold change) of transcripts specified by biotype. Color-matched points along the x axis represent median values for each dataset. Number of transcripts per group (n) and p values compared with “all” group generated via Kolmogorov–Smirnov test are shown in bottom right and matched to group color. Light orange represents annotated NMD transcripts. Black represents all transcripts. Light pink represents annotated protein-coding transcripts. C, CDF, as described above, of axonal abundance ratio of transcripts specified as upregulated (orange) or downregulated (magenta). D, CDF, as described above, of 3′ UTR length normalized UPF1 binding elements of transcripts specified as upregulated (orange) or downregulated (magenta). E, CDF, as described above, representing length of transcripts specified by differential expression, upregulated (orange) or downregulated (magenta). F, CDF, as described above, of 3′ UTR length normalized to overall transcript length of transcripts specified by differential expression, upregulated (orange) or downregulated (magenta). G, CDF, as described above, of coding sequence length normalized minimum free energy (–ΔG) of transcripts specified as upregulated (orange) or downregulated (magenta). H, CDF, as described above, of 3′ UTR length normalized minimum free energy (–ΔG) of transcripts specified as upregulated (orange) or downregulated (magenta). I, Variable importance plot showing the level of importance each variable discussed plays in determining fold change after 11j treatment. Variables are ordered from high importance (top, pink) to low importance (bottom, orange). Bioinformatic analyses of the 5′UTR can be found in Extended Data Figure 3-1.
Figure 3-1
Analysis of high-throughput sequencing reveals unique features in the 5′ UTR of NMD targets in the DRG. A, CDF of 5′ UTR length normalized to overall transcript length of transcripts specified by differential expression, upregulated (orange) or downregulated (magenta). Color-matched points along the x axis represent median values for each dataset. Number of transcripts per group (n) and p values as compared to 'all' group generated via Kolmogorov–Smirnov test are shown in bottom right and matched to group color. B, Bar chart representing the number of transcripts (y axis) in each differential expression group and the number of uORFs they contain (x axis). All transcripts found is represented by the tan bar, while upregulated transcripts are shown in orange and downregulated transcripts are shown in pink. C, CDF as described above of 5′ UTR length normalized minimum free energy (–ΔG) of transcripts specified as upregulated (orange) or downregulated (magenta). Download Figure 3-1, EPS file.
To determine whether previously defined NMD targets are stabilized by SMG1 inhibition in sensory neurons, we calculated their fold change. A major element of our informatics approach consists of cumulative distribution functions (CDFs). In these graphs, the distribution of data between multiple datasets can be compared. We also plot the median values as dots along the x axis. mRNAs were divided into three groups (all mRNAs, protein-coding mRNAs, and annotated NMD targets) and their fold changes were compared. Annotated NMD targets were upregulated after NMD inhibition in agreement with our prior results confirming the effects of 11j on known targets (Fig. 3B). Next, we capitalized on compartmentalized sequencing of DRG neurons to ask whether the distribution of NMD targets was biased toward the soma or axons (Minis et al., 2014). We calculated a ratio to describe the polarity of each mRNA (Minis et al., 2014). A value of 0 indicates a cell body-localized mRNA, while a value of 1 indicates a transcript found exclusively in axons (Minis et al., 2014). In our data, we find that the upregulated transcripts have a small but significant bias toward axonal localization (Fig. 3C). It is unclear whether this bias is caused by destabilization of somatic transcripts or preferential destabilization of localized mRNAs by NMD.
NMD targets have been identified in cell lines based on stability following knockdown of UPF1 and physical interactions between UPF1 and mRNA targets (Imamachi et al., 2017). Intriguingly, a cis-acting element has been reported in the 3′ UTR of UPF1 targets. This consists of the following G-rich consensus element: CCTG[GA][GA][GA] (Imamachi et al., 2017). We found that this element was strongly enriched in our NMD target set and depleted in mRNAs that were reduced in abundance following 11j treatment (Fig. 3D). This suggests that signatures of UPF1 binding are present in our NMD target set obtained through SMG1 inhibition.
NMD sensitivity has been linked to UTR length. One model posits that mRNAs that possess extended 3′ UTRs are subject to NMD (Kebaara and Atkin, 2009). In contrast to these findings, we found that DRG NMD targets had shorter 3′ UTRs and were generally more compact (Fig. 3E,F). However, the 5′ UTRs of NMD targets were slightly longer (Extended Data Fig. 3-1A), although the overall contribution of this difference to total transcript length is likely miniscule as 3′ UTRs tend to be much longer than 5′ UTRs. We next asked whether additional features present in the 5′ UTRs might contribute to NMD sensitivity.
uORFs are short coding segments upstream of translational start sites. uORFs have been linked to NMD as the uORF stop codon can be recognized as a PTC (Barbosa et al., 2013). We found that there was no significant enrichment of transcripts harboring uORFs as defined by ribosome profiling studies among NMD targets (Extended Data Fig. 3-1B) (Ingolia et al., 2011; Johnstone et al., 2016; Barragán-Iglesias et al., 2021). Additionally, NMD targets did not possess a disproportionate number of transcripts with large numbers of uORFs. mRNAs with >5 uORFs were slightly more abundant in transcripts that decreased on 11j treatment.
Finally, one model for NMD posits that RNA structural content is a major determinant (Fischer et al., 2020). Therefore, we aimed to predict the stability of different regions of mRNAs sensitive to SMG1 inhibition. The RNAfold program from ViennaRNA takes an input sequence of nucleotides, calculates its minimum free energy, and uses dynamic programming methods to predict the secondary structure of a sequence (Lorenz et al., 2011). The minimum free energy output is multiplied by −1 and normalized to the length of the sequence. This value is a measurement of predicted structural content (–ΔG/nt). To investigate whether structure content influenced NMD, we asked whether NMD targets possess a higher degree of structure than nontargets. Indeed, structural density in both the coding sequence and the 3′ UTR was significantly increased in NMD targets while it had very little effect in the 5′ UTR (Fig. 3G,H; Extended Data Fig. 3-1C). These data indicate that structure, primarily in the 3′ UTR, is a defining feature of stability control by NMD in DRG neurons. But which feature is the most important?
To compare the relative predictive power of each of the aforementioned parameters, we calculated random forest-based importance. The random forest model is a supervised machine learning algorithm in which the algorithm learns the relationship between variables. To do this, multiple decision trees are merged, ensuring an accurate prediction model. We wanted to test how each characteristic examined (e.g., transcript length, structure content, etc.) was predictive of fold change. We found that 3′ UTR structure was the most important determinant (Fig. 3I). This is notable as the relationship between structural content in the 3′ UTR has only been detected in somatic cells (Fischer et al., 2020). Additionally, features that have been reported in somatic cells as major determinants of NMD, such as uORFs and 5′ UTR length, are relatively unimportant in the DRG. This result underscores the need for detailed characterization of post-transcriptional control pathways in the nervous system as they may differ from somatic cell types.
Inhibition of NMD induces mechanical and thermal hypersensitivity in mice
We next asked whether inhibition of NMD via SMG1 induces pain-associated behaviors in mice. A common approach to quantify allodynia involves measurement of mechanical withdrawal thresholds using calibrated von Frey filaments. Filaments of various thicknesses are pressed against the skin of the hind paw until paw flinching is observed. Enough force is applied to induce bending of the filament. Because the amount of force required to bend the filament is dependent on the filament's gauge, the amount of stimulation required to elicit a reflexive behavioral response (paw withdrawal) can be precisely quantified. The amount of force required to elicit paw withdrawal in a naive animal is ∼1 g. Noxious cues, such as inflammatory mediators, result in an increased sensitivity to formerly innocuous stimuli. This primed state is commonly evoked with a subthreshold dose of a mild stimulus (Reichling and Levine, 2009). We injected 11j (1 or 6 µg) into the hind paw and scored mechanical hypersensitivity. We found that 11j induces a significant decrease in withdrawal threshold (allodynia) that lasts for 3 d before returning to the baseline level of sensitivity (Fig. 4A). After the animals returned to baseline, we tested if they were still primed using a subthreshold dose (100 ng) of prostaglandin E2 (PGE2). This mild dose of PGE2 does not induce sensitivity in vehicle-treated mice. However, it evoked significant mechanical hypersensitivity in mice who previously received 11j (Fig. 4B). This phenomenon is called hyperalgesic priming (Reichling and Levine, 2009). In a complementary set of experiments, we measured thermal sensitivities using the Hargreaves test. We found that 11j increased thermal sensitivity both acutely (3 h after 11j injection) (Fig. 4C) and resulted in a primed state evoked days later by a subthreshold dose of PGE2 (3 h after PGE2 challenge) (Fig. 4D). In females, 11j resulted in mechanical hypersensitivity but not thermal hypersensitivity (Extended Data Fig. 5-1A–D). These findings imply that NMD plays a role in nociceptive processing and that inhibition of NMD in the periphery induces mechanical sensitization in mice. The effects of NMD inhibition on thermal sensitization are sexually dimorphic.
Inhibition of NMD induces mechanical and thermal hypersensitivity in mice. A, Intraplantar (i. pl.) injection of 11j decreased the withdrawal threshold of mice to von Frey filaments, indicative of mechanical hypersensitivity. Two-way repeated-measures ANOVA showed treatment (F(2,15) = 36.26, p < 0.0001) and time (F(3.775, 56.63) = 15.86, p < 0.0001) effects. *p < 0.05; **p < 0.01; ***p < 0.001; versus the vehicle group (Dunnett's multiple comparisons test). B, Following the resolution of the acute mechanical hypersensitivity, on day 9, mice were challenged with a subthreshold dose of PGE2 (100 ng, i. pl.). Mice treated with 11j exhibited decreased withdrawal threshold indicating the development of hyperalgesic priming. Two-way repeated-measures ANOVA revealed treatment (F(2,15) = 16.49, p < 0.001) and time (F(1.737, 26.06) = 7.348, p < 0.01) effects. *p < 0.05; **p < 0.01; ***p < 0.001; versus the vehicle group (Dunnett's multiple comparisons test). C, 11j also reduced withdrawal latency on application of a radiant heat source (Hargreave's method), indicating thermal hypersensitivity. One-way ANOVA demonstrates treatment effect (F(2,15) = 16.70, p < 0.001). ***p < 0.001 versus the vehicle group (Dunnett's multiple comparisons test). D, This thermal hypersensitivity was also observed after PGE2 challenge, indicating the development of hyperalgesic priming. One-way ANOVA revealed treatment effect (F(2,15) = 10.13, p < 0.01). **p < 0.01 versus the vehicle group (Dunnett's multiple comparisons test). Data are mean ± SEM; n = 6 animals per group.
ISRIB rescues the pro-nociceptive effects of 11j
We next asked why inhibition of NMD is pronociceptive. Given that ATF4 is an NMD target and that induction of ATF4 is a hallmark of the ISR, we asked whether blockade of NMD induces the ISR (Wengrod et al., 2013). Phosphorylation of eIF2α is a marker for induction of the ISR. We found that 11j treatment (1 μm, 4 h) led to a slight but significant increase in eIF2α phosphorylation (Fig. 5A). Importantly, the abundance of eIF2α was not influenced by 11j. We also examined translation of ATF4 following addition of 11j. Induction of ATF4 by 11j was blunted by addition of ISRIB (Fig. 5B). To probe the underlying mechanism, we examined the list of transcripts impacted by 11j. Intriguingly, the transcript that encodes CReP, PPP1R15b, was reduced by 34%. We examined accumulation of CReP using immunocytochemistry. We found that, specifically in the population of sensory neurons that express peripherin, CReP was reduced by 37% (Fig. 5C). Collectively, the data suggest that 11j results in increased eIF2α phosphorylation and a reduction in the corresponding phosphatase CReP.
Pharmacological inhibition of the ISR rescues the pro-nociceptive effects of NMD inhibition. A, 11j treatment induced eIF2α phosphorylation (p-eIF2α), which was reversed by cotreatment of ISRIB. Left, Representative images. Right, Quantification of p-eIF2α/eIF2α levels in cultured DRG neurons. n = 4 biological replicates per treatment group. *p < 0.05 (two-tailed unpaired t test). B, 11j treatment also induced ATF4 expression, which was reversed by cotreatment of ISRIB. Left, Representative images. Right, Quantification of ATF4/GAPDH levels in cultured DRG neurons. n = 4 biological replicates per treatment group. One-way ANOVA revealed treatment effect (F(2, 9) = 58.39, p < 0.001), and Tukey's multiple comparisons test showed difference between Vehicle versus 11j (***p < 0.001) and 11j versus 11j + ISRIB (***p < 0.001). C, 11j treatment led to a 37% reduction in CReP immunofluorescence in cultured DRG neurons. Left, Representative immunofluorescence images. Scale bar, 20 µm. Right, Quantification of CReP expression by CTCF. Welch's t test (unpaired, two tailed) showed a difference between Vehicle (n = 78) versus 11j (n = 66), ***p < 0.001. D, ISRIB ameliorated the acute mechanical hypersensitivity induced by 11j. Two-way repeated-measures ANOVA showed treatment (F(2, 15) = 28.98, p < 0.0001) and time (F(3.302, 49.54) = 5.189, p < 0.01) effects, and Dunnett's multiple comparisons test revealed significant difference versus the 11j-treated group. *p < 0.05. **p < 0.01. ***p < 0.001. E, ISRIB also prevented the mechanical hypersensitivity induced by 11j during the priming phase of the hyperalgesic priming model. Two-way repeated-measures ANOVA demonstrated treatment (F(2,15) = 8.719, p < 0.01) and time (F(1.963, 29.44) = 5.185, p < 0.05) effects, and Dunnett's multiple comparisons test indicated significant difference versus the 11j-treated group. *p < 0.05. **p < 0.01. ***p < 0.001. F, ISRIB also reduced the acute thermal hypersensitivity induced by 11j. One-way ANOVA revealed treatment effect (F(2,15) = 6.685, p < 0.01), and Dunnett's multiple comparisons test confirmed significant difference versus the 11j-treated group. *p < 0.05. **p < 0.01. G, This amelioration of 11j-induced thermal hypersensitivity by ISIRB was also during the priming phase. One-way ANOVA demonstrated treatment effect (F(2,15) = 7.286, p < 0.01), and Dunnett's multiple comparisons test indicated significant difference versus the 11j-treated group. *p < 0.05. **p < 0.01. Data are mean ± SEM; n = 6 animals per group. Behavioral tests in female mice can be found in Extended Data Figure 5-1.
Figure 5-1
The nociceptive effects of 11j and its attenuation by ISRIB were also observed in female mice. A, Intraplantar (i.pl.) injection of 11j decreased the withdrawal threshold of female mice to von Frey filaments, indicating mechanical hypersensitivity, which was rescued by cotreatment with ISRIB. Two-way repeated-measures ANOVA showed treatment (F(2, 15) = 9.024, p < 0.01) and time (F(3.261, 48.92) = 6.668, p < 0.001) effects, and Tukey's multiple comparisons test revealed significant difference versus the Vehicle group *p < 0.05, **p < 0.01 or versus the 11j-treated group #p < 0.05, ##p < 0.01. B, Female mice previously treated with 11j exhibited decreased withdrawal threshold after PGE2 injection, indicating the development of hyperalgesic priming. This was not observed in female mice who received cotreatment of 11j and ISRIB. Two-way repeated-measures ANOVA showed treatment (F(2, 15) = 8.484, p < 0.01) and time (F(1.820, 27.30) = 6.668, p < 0.01) effects, and Tukey's multiple comparisons test revealed significant difference versus the Vehicle group *p < 0.05 or versus the 11j-treated group #p < 0.05. C, D, No significant effects were observed in thermal sensitivity in both the acute (C) and priming phases (D) of the hyperalgesic priming model. Data are means ± SEM. n = 6 animals per group. Download Figure 5-1, EPS file.
Next, we asked whether the nociceptive effects of 11j were affected by a small-molecule inhibitor of the ISR. ISRIB acts downstream of eIF2α phosphorylation to restore translational homeostasis (Schoof et al., 2021; Zyryanova et al., 2021). We found that cotreatment with ISRIB (1 or 10 µg) rescued the mechanical hypersensitivity induced by 11j in mice, in both the acute phase (Fig. 5D) and after priming (Fig. 5E). ISRIB also blocked the thermal hypersensitivity induced by 11j at both 3 h after 11j injection (Fig. 5F) and 3 h after PGE2 challenge (Fig. 5G). Furthermore, the nociceptive effects of 11j and its attenuation by ISRIB were also observed in female mice (Extended Data Fig. 5-1A,B). Collectively, our results suggest that the impact of NMD on nociception likely occurs through stimulation of the ISR.
Discussion
The data reveal the repertoire of mRNAs whose abundance is controlled by NMD. Parallel experiments enable three major conclusions. First, NMD targets in sensory neurons possess highly structured 3′ UTRs. Second, blockade of NMD results in activation of the ISR. Third, transient disruption of NMD results in mechanical hypersensitivity that is reversed by inhibition of the ISR. We discuss the implications of each in turn.
A variety of pathways implicated in pain regulate RNA stability. For instance, the poly(A) binding protein promotes RNA stability and translation (Burgess and Gray, 2010). Inhibition of poly(A) binding proteins with a synthetic poly(A) decoy results in diminished behavioral responses to both inflammatory mediators and a postsurgical pain model involving incision (Barragán-Iglesias et al., 2018). The cytoplasmic polyadenylation element-binding is activated by CaMKIIα and controls cytoplasmic polyadenylation (Atkins et al., 2004). Knockdown of cytoplasmic polyadenylation element-binding prevents carrageenan-induced priming, suggestive of an important role in neuroplastic changes that contribute to persistent effects of inflammatory mediators on nociceptors (Bogen et al., 2012). miRNAs have also emerged as key players in pain signaling. Conditional deletion of the molecule required for the maturation of most miRNA species, DICER, from sensory neurons that express Nav1.8 results in profound alterations in pain thresholds (Zhao et al., 2010). Similar to both cytoplasmic polyadenylation element-binding and poly(A) binding proteins, miRNAs regulate both translation and RNA stability. Precise dissection of stability apart from translation is complicated by the intimate connectivity of the two processes. For example, if stability is altered, translation will likely be affected. Nonetheless, many of these pathways converge on regulation that modulates poly(A) tail length. An additional commonality is the use of information present in the 3′ UTR. A key finding of our study is that the most prominent correlation among NMD targets is the extent of structure in this region. What is the trans-acting factor that facilitates recognition of structured elements and destabilization through NMD? The most parsimonious factor is G3BP1. G3BP1 is commonly used as a marker for stress granules, a product of eIF2α phosphorylation (Aulas et al., 2015; Sidrauski et al., 2015). In cell lines, 3′ UTR structure-mediated decay requires both G3BP1 and UPF1 (Fischer et al., 2020). G3BP1 is present in DRG sensory neurons (Sahoo et al., 2018). A key question is whether structure-mediated decay is mechanistically similar in the nervous system and whether phase separation by G3BP1 in response to the ISR influences NMD activity on these substrates.
NMD regulates the stability of a large segment of the transcriptome. To date, a single NMD target has been described in the DRG. Following spared nerve injury, a specific variant of versican accumulates that possesses two PTCs (Bogen et al., 2019). This likely triggers NMD. Why is NMD suppressed during spared nerve injury? NMD requires ongoing translation, and eIF2α is phosphorylated during spared nerve injury. Thus, translational inhibition by activation of the ISR likely contributes to blockade of NMD as has been reported in hypoxic cells (D. Wang et al., 2011). Our data suggest that loss of NMD likely potentiates the ISR. Notably, we validated ATF4 as a target of NMD in DRG neurons. ATF4 is the main effector of the ISR and mediates cellular adaptation to stress and survival (Lu et al., 2004; Pakos-Zebrucka et al., 2016). It also contributes to thermal but not mechanical hypersensitivity (Xie et al., 2021). We speculate that the convergence of NMD and the ISR on ATF4 might provide a means to ensure regulatory fidelity during stress.
Does activation of NMD prevent neuroplastic changes induced by activation of the ISR? While we do not know the answer, several enhancers of NMD have been reported. They include Tranilast, mitoxantrone hydrochloride, and trazodone hydrochloride, among others (Xu et al., 2019; Zhao et al., 2022). The exact mechanism by which these drugs modulate NMD is still unclear. We found that these drugs do not increase UPF1 phosphorylation in DRG neurons (data not shown). This suggests that they likely act on branch-specific pathways as opposed to the SMG1-UPF1 axis. A key question moving forward is whether agonists of SMG1 have an opposing effect on pain driven by activation of the ISR.
NMD modulation has emerged as a potential strategy for an array of disease states. Activation of the pathway has been proposed as treatment for neurodegenerative diseases, such as amyotrophic lateral sclerosis and frontotemporal dementia (Barmada et al., 2015; Jackson et al., 2015; Tank et al., 2018; Xu et al., 2019). However, in cancer, NMD plays more idiosyncratic roles (Nogueira et al., 2021). In lung adenocarcinoma and hepatocellular carcinoma, NMD activation limits proliferation, while in hereditary diffuse gastric cancer and colorectal cancers with microsatellite instability, NMD inhibition restricts cell growth (Karam et al., 2008; Chang et al., 2016; Cao et al., 2017; Bokhari et al., 2018). Additionally, NMD inhibition promotes translational readthrough and has long been investigated for disease states caused by nonsense mutations, such as cystic fibrosis, myotonic dystrophy, and β-thalassemia (Lai et al., 2004; Salvatori et al., 2018; Borgatti et al., 2020; Erwood et al., 2020; Sanderlin et al., 2022). Our work demonstrates that suppression of NMD promotes nociceptive sensitization. This can be mitigated by inhibition of the ISR. Together, these observations may have important implications for approaches aimed at suppression of NMD in other disease states.
One limitation of our study was the use of steady-state measurements of mRNA abundance. We assayed the transcriptome using a method that captures the 3′ end of mRNA, QuantSeq. Our approach captures both direct and indirect effects of NMD inhibition at the resolution of reads per gene. The intent of the experiment was to determine how NMD inhibition impacts the transcriptome. Yet, NMD has preferential effects on specific isoforms. There are more sensitive approaches that capture stability and transcript diversity. For example, RNA stability can be more accurately quantified with pulse-chase sequencing methods (Paulsen et al., 2014; Herzog et al., 2017; Wolfe et al., 2020). Long-read sequencing is an emerging tool for detection of isoform-specific effects (Karousis et al., 2021). Despite these limitations, we were able to capture annotated NMD targets (Fig. 3B).
In conclusion, our data suggest that NMD regulates nociceptive signaling through effects on the ISR, notably reduction of CReP. This information is useful for the design of experiments that target NMD in a broad array of biological contexts. We uncover a link between dominant regulatory pathways with implications on deciphering coordinate regulation of mRNA networks.
Footnotes
This work was supported by National Institutes of Health Grant R01NS114018 to Z.T.C., Grant R24GM141526 to S.E.B., and Grant R01NS065926 to T.J.P.
The authors declare no competing financial interests.
- Correspondence should be addressed to Zachary T. Campbell at zcampbell{at}wisc.edu