Human DEAH-box helicase 8 regulates HSF1-mediated stress response and cancer-associated pre-mRNA splicing in tumour cells
Jennifer R Tall, Robert te Poele, Alexandra Vasile, Pradeep Ramagiri, James Campbell, Caitlin R Davies, Marissa V Powers, Toby Roe, Deivendran Sankaran, Hannah Wang, Konstantinos Mitsopoulos, Bissan Al-Lazikani, Rob L M van Montfort, Emmanuel de Billy, Paul Workman, Paul A Clarke
TL;DR
This study shows that DHX8, an RNA helicase, regulates HSF1 and cancer-related gene splicing, making it a potential target for cancer therapy.
Contribution
DHX8 is identified as a novel regulator of HSF1 and cancer-associated pre-mRNA splicing, with potential therapeutic implications.
Findings
DHX8 silencing reduces HSF1 protein by causing intron retention in its transcripts.
Loss of DHX8 alters RNA processing of cancer-associated genes linked to poor outcomes.
DHX8 silencing induces apoptosis more effectively in cancer than in non-tumorigenic cells.
Abstract
Transcription factor heat shock factor 1 (HSF1) orchestrates the cellular stress response, promoting malignant transformation, unchecked proliferation, and stress-resilient survival of tumour cells. We set out to discover potentially druggable regulators of HSF1 activation and identified DEAH-box RNA helicase 8 (DHX8). We investigated the role of DHX8 in regulating HSF1 within the broader context of DHX8 function in cancer cells. DHX8 silencing induces intron retention in HSF1 transcripts, reducing HSF1 protein. Importantly, DHX8 loss significantly alters RNA processing of an HSF1-regulated cancer-associated gene signature linked to poor clinical outcomes, as well as additional oncogenic and stress-response pathways. DHX8 binds between the pre-messenger RNA (mRNA) lariat branch point and the 3′ splice site, consistent with the predominance of intron-retained transcripts following DHX8…
Genes, proteins, chemicals, diseases, species, mutations and cell lines named across the full text — each resolved to its canonical identifier and authoritative record.
Click any figure to enlarge with its caption.
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
Figure 7- —Cancer Research UK10.13039/501100000289
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsRNA Research and Splicing · Heat shock proteins research · RNA regulation and disease
Introduction
Cancer cells encounter various stressors throughout their lifetime, including oncogene activation, nutrient deprivation, DNA damage, and proteotoxic stress [1–4]. Targeting the proteins involved in pathways that enable malignant cells to withstand these stresses represents a promising avenue for therapeutic intervention [5–7]. Protection against proteotoxic stress is primarily mediated by molecular chaperones, whose canonical expression is increased through activation of the transcription factor HSF1 (heat shock factor 1) [1, 3, 4, 8–10]. These chaperone-mediated protective mechanisms can attenuate the effectiveness of therapies designed to induce proteotoxic stress, such as HSP90 or proteasome inhibitors [4, 7, 8, 11, 12].
In addition to regulating canonical heat shock genes, HSF1 controls a broad transcriptional program encompassing genes encoding proteins essential for cell growth and survival, and involved in processes such as protein synthesis, cell cycle regulation, and glucose metabolism [4, 8, 13]. HSF1 also facilitates oncogenic transformation and HSF1-deficient mice exhibit greater resistance to tumour development [14]. The heightened reliance of cancer cells on HSF1 is thought to stem from both the heavy burden placed on the chaperone machinery by mutated or highly expressed oncoproteins and the need for cytoprotective factors to cope with the stressful tumour microenvironment. Overexpression of HSF1 has been observed in various cancer types, and increased HSF1 activity along with its cancer-associated transcriptional signature is associated with poor clinical outcomes [3, 8, 15, 16].
Given the strong link between HSF1 and malignancy, there is considerable interest in developing small-molecule inhibitors of HSF1 for cancer therapy [6, 7, 17]. However, as a ligandless transcription factor, HSF1 remains a notoriously challenging drug target, as evidenced by the limited number of available inhibitors, most of which lack specificity [17]. To overcome this limitation, we used a small interfering (siRNA) screening approach to identify regulators of the HSF1 stress response as potential alternative therapeutic targets to suppress HSF1 activity and identified DEAH-box RNA helicase 8 (DHX8), the human homologue of the yeast Prp22 RNA helicase. At the time of this discovery, little was known about the role of the human DHX8 in mammalian cells. Therefore, we further explored the function of DHX8 in human cancer cells, focusing on HSF1 regulation and employing global and orthogonal approaches to better define its broader cellular functions and the mechanisms underlying HSF1-regulated gene expression within these contexts. Overall, our study identifies DHX8 as a critical regulator of HSF1 messenger RNA (mRNA) processing and the subsequent induction of stress-response gene expression, as well as broader effects, and suggests DHX8 as a potential cancer therapeutic target.
Materials and methods
Materials
The HSP90 inhibitor, 17-AAG (tanespimicin), was purchased from Sigma–Aldrich.
Cell culture
Human cell lines were cultured in DMEM (Sigma–Aldrich) supplemented with 10% foetal calf serum (P.A.A. Laboratories), 2 mM glutamine, and non-essential amino acids (Thermo Fisher Scientific). Cell identity was confirmed using short tandem repeat profiling, and all cell lines tested negative for Mycoplasma.
Generation of cell line models
HCT116 doxycycline-inducible DHX8 model
Stable tetracycline repressor expressing human colon cancer HCT116 T-REx cell lines were generated using the pLenti3.3/TR vector (Thermo Fisher Scientific). QuickChange Lightning site-directed mutagenesis (Agilent Technologies) was used to generate DHX8 mutants. DHX8 wild-type (WT) and K594E, R647G-F648G, or Y813G-S814G-A815G mutant constructs were cloned into the pLenti6.3/DEST (Thermo Fisher Scientific) expression vector via Gateway recombination from pENTR4/FLAG-HA DHX8 WT or mutant entry clones to generate doxycycline-inducible tagged-DHX8 expression constructs. Lentiviral particles were produced in 293T/17 cells using second-generation packaging vectors psPAX2 and pMD2.G. Viral supernatants containing FLAG-HA tagged-DHX8 constructs were used to transduce the HCT116 T-REx cells. Transformed cells were expanded and stable clonal cell lines established. For rescue experiments, stable clones were treated ± induction with 1 μg/ml doxycycline for 48 h. Cells were then transfected using Lipofectamine RNAiMAX (Thermo Fisher Scientific) with 50 nM siRNAs (Supplementary Table S1) for 72 h and cells harvested for analysis.
DHX8–dTAG H1299 NSCLC cells
A full-length human DHX8 open reading frame was synthesized, incorporating an in-frame N-terminal smHiBit tag for expression detection using a HiBiT luminescence (Promega), followed by a FKBP12^F36V^ degron, and cloned into the pDONR221 entry vector (Invitrogen). The DHX8–dTAG gene was sub-cloned into plasmid F560 using Gateway recombination (Invitrogen), a derivative of pEFIRES-P [18]. The construct was transfected into H1299 non-small cell lung cancer (NSCLC) cells and stable pools were selected using 5 ug/ml puromycin. To eliminate endogenous DHX8 expression, CRISPR/Cas9 genome editing was performed using single guide RNAs (sgRNA) that selectively targeted the endogenous DHX8 gene. Single-cell clones were isolated using serial dilution. Editing and loss of endogenous DHX8 was confirmed by sequencing and immunoblotting.
siRNA transfection
U2OS human osteosarcoma cells were reverse-transfected with 50 nM human Druggable Genome SMARTpool siRNA library (Dharmacon) or individual siRNAs (Dharmacon or Qiagen) using HiPerFect transfection reagent (Qiagen). We selected 72 h post-transfection as the primary assay time point, consistent with standard practice in cancer cell siRNA viability screens and with guidance that protein depletion and phenotypic effects typically lag mRNA knockdown (48–96 h), making ~72 h a robust screening window recommended by the library manufacturer (Dharmacon). Scrambled negative control or AllStar negative control siRNAs were from Dharmacon or Qiagen, respectively. The other cell lines used in this study were reverse-transfected with 50nM siRNA using Oligofectamine (Thermo Fisher Scientific). For RNA-seq profiling, H1299 cells were reverse-transfected with 50 nM pooled DHX8 siRNA (siTOOLS Biotech) using Lipofectamine RNAiMAX (Thermo Fisher Scientific).
Cell viability, cell cycle, and apoptosis assays
Cell-TiterGlo and Caspase 3/7-Glo assays (Promega) were performed according to the manufacturer’s protocol and read using an EnVision plate reader (Perkin Elmer). Cell cycle analysis was carried out as previously described [11].
Protein expression
Protein expression was determined using immunofluorescence or gel-based polyacrylamide immunoblotting [6, 11], or using a capillary immunoassay (Protein Simple).
Primary antibodies used were: HSF1 (ADI-SPA-901, RRID:AB_10616511), HSP27 (ADI-SPA-800, RRID:AB_10616382), and heat shock protein 72 (HSP72) (ADI-SPA-810, RRID:AB_10616513) from Enzo Life Sciences; DHX8 (ab54592, RRID:AB_941290, or ab181074) or HA-tag (ab72479, RRID:AB_10711040) from Abcam; cleaved PARP (9541, RRID:AB_331426), total PARP (9542, RRID:AB_2160739), and cleaved caspase 3 (9664, RRID:AB_2070042) from Cell Signaling Technology; and GAPDH (MAB374, RRID:AB_2107445) from EMD Millipore. Cells were harvested by scraping in phosphate-buffered saline (PBS) and lysed in cell lysis buffer (0.05M Tris, pH7.6, 0.15M NaCl, 0.5% NP40, and Complete Protease Inhibitor Cocktail; Roche) on ice. The cell lysate was centrifuged at 17 000 × g for 15 min at 4°C and the supernatant recovered. Protein concentrations were determined using the Pierce™ BCA Protein Assay Kit.
The capillary immunoassay used protein lysates loaded onto the 12–230 kDa (Bio-Techne) assay plate. Secondary antibodies and detection reagents were obtained from detection packs for anti-rabbit or mouse, respectively (Protein Simple). Standard default capillary electrophoresis and detection conditions were used. Data were analysed using the Compass software (Protein Simple).
For immunofluorescence, cells were fixed in 4% paraformaldehyde, permeabilized with 0.5% NP40. Expression was detected using HSF1 or HSP72 primary antibodies and Alexa Fluor^®^ 488 goat anti-mouse IgG secondary antibodies (GE Healthcare). Cells were counterstained with DAPI (Invitrogen) and were imaged and fluorescence quantified using an INCell Analyzer 1000 (GE Healthcare) and INCell Investigator software.
RNA analysis or profiling
TaqMan® assays
TaqMan assay primers were obtained from Applied Biosystems: HSPA1A (Hs00359163_s1), HSPB1 (Hs03044127_g1), HSF1 (Hs00232134_m1), un-spliced HSF1 (Hs03673241_cn), spliced NOXA (Hs00560402_m1), un-spliced NOXA (Hs00906409_cn), spliced MYC (Hs0153408_m1), CDC37 (Hs01003386_g1), and control RPLP0 (4326314E). For HSF1 splicing experiments, additional primers were custom-made for spliced HSF1 overlapping exon1–exon2 (forward: ACCGACGCGCTCATCTG and reverse: GGTCGAACACGTGGAAGCT) to pair with the un-spliced HSF1 primers (above) measuring intron1.
The cells were lysed in Cells-to-cDNA II cell lysis buffer (Life Technologies) and transferred to polymerase chain reaction (PCR) plates. Complementary DNA (cDNA) synthesis was performed using the High-Capacity cDNA Reverse Transcription Kit (Life Technologies). TaqMan reactions were performed in 384-well plates using 1× FAM- and VIC-labelled primer probe sets, and TaqMan Universal Master Mix (Life Technologies). Plates were run for 40 cycles using an ABI Prism 7900HT Sequence Detection System and analysed using SDS 2.1 software.
Microarray analysis
RNA was extracted using the ABI PRISM 6100 Nucleic Acid PrepStation (Applied Biosystems). Samples were analysed using Agilent’s two-colour microarray-based gene expression analysis with a Universal Human Reference RNA (Stratagene) [19].
Background-subtracted data for the sample and reference were analysed using Genespring analysis software (Agilent). Data were Lowess normalized and probes showing >1.5-fold altered expression, with a corrected P-value <.05, were deemed altered or enriched. In the RNA immunoprecipitation (RIP) experiment, probes with at least >1.5-fold enrichment in the DHX8 IP above the IgG control were included in the analysis.
ANOVA and moderated t-test were used to determine significance and P-values were adjusted for multiple testing using the Benjamini–Hochberg correction with a false discovery rate of 5%. Potential relationships between the significant genes and pathways and transcriptional regulation were investigated using Metacore software (GeneGo).
RIP assay
The RIP assay was performed in U2OS cells using the EZ-Magna RIP RNA-Binding Protein Immunoprecipitation Kit (Millipore), according to the manufacturer’s instructions. Five micrograms of DHX8 antibody (ab54592; Abcam) was used for immunoprecipitation.
RNA-seq
Total RNA was extracted using the MagNa Pure 96 automated platform (Roche). Sequencing libraries were prepared (MGIEasy Fast RNA Library Prep Set) following ribosomal RNA depletion (MGIEasy rRNA Depletion Kit). Libraries were sequenced on a BGISEQ-500 or Illumina platform using strand-specific, paired end 100 bp read length, and 50M clean reads were collected.
FastQ files were aligned to the reference genome (GRCh38) using the nf-core/RNA-seq pipeline (version 21.04.3) comprizing the following tools: Read QC (FastQC v0.11.9 and MultiQC v1.12), UMI extraction (UMI-tools), adapter, and quality trimming (Trim Galore!), removal of genome contaminants (BBSplit), alignment to the reference genome (STAR v.2.7.6a), sorting and index alignments (SAMtools), duplicate read marking (Picard MarkDuplicates), and bigWig coverage files (BEDTools, BedGraphToBigWig). HTSeq-count (HTSeq v0.12.4) was used to count the number of unambiguously mapped reads. Differential expression analysis was performed using the Bioconductor package DESeq2 (v1.20.0). Pathway and gene set enrichment analyses were performed using the fgsea package (v1.0).
The RNA-seq data were generated from short-read sequencing, which cannot reliably be resolved to analyse different alternatively spliced full-length transcript isoforms from an individual gene; therefore, we limited our analysis to canonical splice-junction events. Accordingly, gene expression and splice-junction usage were analysed as independent but complementary metrics, consistent with best practices for short-read RNA-seq splicing analysis [20, 21]. Differential splicing analysis was performed by creating first pass and second pass BAM files by mapping raw reads to the reference genome (GRCh38) using STAR (v.2.7.6a). U2OS data were analysed with SplAdder and SplAdder-test (https://github.com/ratschlab/spladder/) using the development branch commit (e92ec4487b39c71f0f5f7d25916c175b0b7864a2). All other splicing analyses used differential local splicing variations detected, quantified using MAJIQ (v2.2-e25c4ac), and visualized using Voila (v2.2.0-e25c4ac) software packages.
eCLIP
Enhanced crosslinking and immunoprecipitation experiments were performed under contract by EclipseBio using a single-end SeCLIP protocol [22]. HCT116 cells were labelled with 200 µM 4-thiouridine for 8 h followed by irradiation on ice with UV light at 365 nm (2000 µJ/cm^2^) to photocrosslink RNA–protein complexes. Cells were scraped into ice-cold PBS and lysed in proprietary eCLIP lysis buffer (EclipseBio). Lysates were sonicated (Q800R2; QSonica) for 4 min with 30-s on/off cycles. RNA–protein complexes were recovered by immunoprecipitation using 5 µg of rabbit monoclonal anti-DHX8 antibody (ab181074; Abcam), pre-coupled to sheep anti-rabbit IgG Dynabeads (Thermo Fisher Scientific), followed by extensive washing in high-stringency eCLIP wash buffers (EclipseBio). Immunoprecipitated complexes and 1% paired input samples were resolved on NuPAGE 3%–8% Tris–acetate gels (Invitrogen). Proteins were transferred to membranes, and the region spanning 140–215 kDa (encompassing DHX8 and ~75 kDa above its molecular weight) excised for RNA extraction.
Sequencing libraries were prepared following established eCLIP protocols [22]. Raw sequencing reads were trimmed to remove 3′ adapters and high-quality reads were aligned to the human reference genome (hg38) using STAR aligner. Duplicate reads were removed using UMI tools. Peaks associated with DHX8 binding were identified using CLIPper algorithm. A peak was defined as a read cluster with a log_2_-fold enrichment ≥3 and a P-value ≤.001 determined by Fisher’s exact test. Enriched sequence motifs were identified using the HOMER suite to categorize the binding sites according to their genomic features and enriched sequence motifs.
Results
An siRNA screen for regulators of HSPA1A mRNA induction by HSF1
To identify druggable regulators of HSF1 activation in cancer cells, we conducted a focused siRNA library screen targeting 7598 genes encoding proteins predicted to be amenable to small-molecule inhibition. HSF1 activation was evaluated by quantifying HSPA1A mRNA accumulation after HSP90 inhibition by 17-AAG (tanespimicin; Fig. 1A) [23]. We selected U2OS human osteosarcoma cells for our screen because we had previously demonstrated that these cells were well suited to high-throughput phenotypic screening for inhibition of HSF1 activation using HSP1A1 expression as a readout [6, 24, 25]. Expression of HSPA1A, which encodes HSP72, was employed, as its transcription is highly dependent on HSF1 through a well-defined heat shock element in its promoter [26]. The expression of HSPA1A was quantified using duplex TAQman RT-qPCR with normalization to ribosomal protein LRP0 expression. U2OS cells were transfected with the siRNA library or control siRNAs and treated with 250nM 17-AAG on day 3, a concentration that robustly induced HSPA1A expression with acceptable experimental variation (Fig. 1B and Supplementary Fig. S1A).
*An siRNA screen in U2OS human osteosarcoma cells identifies regulators of HSF1-dependent HSPA1A mRNA induction. (A) Schematic overview of the primary siRNA screening workflow. (B) Control condition results from the screen, showing HSPA1A mRNA induction following transfection with control siRNAs and treatment with either DMSO vehicle control or 250 nM of HSP90 inhibitor 17-AAG to activate HSF1. Each data point represents HSPA1A mRNA induction for individual siRNA treatments (n = 2 independent repeats). AllStars Negative Control and HSF1 siRNA were included as negative and positive controls, respectively. The dotted line shows the 60% inhibition cut-off used to define candidate hits for followup. (C) Genes silenced in the screen ranked based on their mean percentage induction of HSPA1A mRNA. Red-shaded highlights gene silencing resulting in HSPA1A mRNA altered by >2-fold (n = 2 independent repeats). The inset plot shows data for the top 10 ranked hits (error bars = data range for n = 2 independent repeats). (D) Induction of HSPA1A mRNA following 250 nM 17-AAG treatment after silencing of candidate hits from the siRNA screen. Each point represents the mean percentage induction of HSPA1A from a single siRNA (n = 3 independent replicates), horizontal bars indicate the overall mean values for the four different siRNA for each target. The grey-shaded region denotes hits that reduced HSPA1A mRNA induction by >60%. Red data points denote hits where ≥2 out of 4 siRNA reduced induction by >60%. (E) Effect of silencing the nine validated hits using four individual siRNAs each measured by RT-qPCR (HSPA1A mRNA) or immunofluorescence (HSP72 protein) following treatment with 250 nM 17-AAG (mean, n = 4). The pink-shaded area indicates siRNAs that reduced HSPA1A mRNA levels by >60%. Induction of (F) HSPA1A mRNA or (G) HSPB1 mRNA following 250 nM 17-AAG treatment after 72-h knockdown of HSF1 or DHX8 using eight distinct siRNAs (mean ± SEM, n ≥ 3). Statistical significance was assessed using one-way ANOVA followed by Dunnett’s multiple comparisons test (ns = not significant, *P < .05, **P < .01, and **P < .005). (B–G) Induction was calculated as a percentage of mean mRNA or protein relative to treatment with the 250 nM 17-AAG and transfection with AllStars negative control siRNA condition (indicated by the black line). (H) Fold decrease in basal HSPA1A mRNA levels following 72-h silencing of DHX8 with two coding sequence (CDS) targeting siRNA (DHX8-O1 and -O3) and HSF1 relative to negative control siRNA (mean ± SEM, n ≥ 3). Only significant differences are indicated on the plots.
The screen showed strong reproducibility between the two independent runs with robust inhibition of HSPA1A mRNA induction by the positive control HSF1 siRNA on each test plate (mean Z′ = 0.51; Spearman r = 0.596, P < .0001; Fig. 1B and Supplementary Fig. S1B and C). siRNA data were ranked by the mean percentage induction of HSPA1A in response to HSF-1 activation (Supplementary Data S1), with 100% induction defined by the mean HSPA1A mRNA level in cells transfected with AllStars negative control siRNA. Silencing 233 genes resulted in a >50% reduction in HSPA1A induction, while knockdown of 239 genes led to a >200% increase (Fig. 1C). Importantly, among the top-ranked library siRNA hits that suppressed HSF1-mediated HSPA1A induction were HSF1, the target of the screen, and HSPA1A and HSPA1B siRNAs, which both target the HSPA1A mRNA readout of the screen (Fig. 1C). These observations provided confidence in the assay’s technical quality and its ability to detect regulators of HSF1-mediated transcriptional activity (Supplementary Data S1).
Validation of siRNA screen hits
A cut-off of >60% inhibition of HSPA1A mRNA induction identified 85 target genes for further validation using inhibition of HSPA1A mRNA and cognate protein induction as endpoints. To eliminate false positives, we re-evaluated these 85 candidate hits using the four individual siRNA sequences from the pooled library. This confirmed nine candidate genes, where silencing caused >60% inhibition of induced HSPA1A mRNA expression with at least two individual siRNAs (Fig. 1D and Supplementary Table S2). Three of the nine genes, EIF2S1, EIF4G1, and RHOB, were subsequently eliminated because of a lack of consistent impact on HSP72 protein induction following 17-AAG mediated HSF1 activation (Fig. 1E).
For the remaining six candidate genes, further validation was performed using eight individual siRNAs per gene. This included four sequences from the siRNA screen and four additional siRNA sequences. We evaluated HSPA1A mRNA expression by RT-qPCR and HSP72 and HSF1 protein levels by immunofluorescence (Supplementary Fig. S2A). Of these, only ALDOA and DHX8 silencing consistently suppressed HSPA1A mRNA expression, with four of eight siRNAs for each gene decreasing expression >60%, while only one siRNA showed no effect (Supplementary Fig. S2B). Although the impact on HSP72 protein levels was less pronounced than on HSPA1A mRNA reduction, silencing either ALDOA or DHX8 consistently decreased HSP72 protein abundance (Supplementary Fig. S2B). Importantly, knockdown of DHX8 led to a robust and consistent depletion of HSF1 protein, with >55% depletion observed in 5/8 siRNA treatments, a more substantial effect than that observed with ALDOA silencing (Supplementary Fig. S2B). Based on these results, DHX8 was selected for further investigation. To our knowledge, these findings provide the first evidence that DHX8 regulates the HSF1-mediated stress response pathway.
DHX8 is required for HSF1 expression and heat shock protein induction
We confirmed the effect of DHX8 silencing on the expression of another HSF1-regulated gene, HSPB1 (encoding HSP27 protein). DHX8 silencing not only inhibited the induction of HSPB1 but also reduced both basal and stress-induced HSPA1A expression (Fig. 1F–H). To further investigate the role of DHX8 in HSF1 regulation, we followed best practices for siRNA experiments and selected two siRNAs (DHX8-O1 and DHX8-O3; Fig. 1F and Supplementary Table S2) [27]. Specifically, we chose the second- and third-ranked siRNAs, as the most active siRNAs from pooled screens can occasionally exhibit disproportionate effects due to off-target activity or non-specific toxicity. The suppression of HSPA1A and HSPB1 mRNA levels in response to HSF1 activation were time-dependent, with maximal inhibition observed 3–4 days after siRNA transfection (Fig. 2A and B). In contrast, HSF1 mRNA levels remained largely unchanged following DHX8 silencing (Fig. 2C), whereas HSF1 protein levels declined 48 h after DHX8 knockdown, and were nearly undetectable after 96 h (Fig. 2D). Importantly, expression of non-HSF1-regulated control genes (NOXA, MYC, and CDC37) were not decreased by the silencing of either DHX8 or HSF1 (Fig. 2E–G), reinforcing the apparent specificity of the DHX8–HSF1 regulatory axis.
*DHX8 knockdown inhibits heat shock gene induction and alters HSF1 splicing in U2OS human osteosarcoma cells. Quantification of heat shock gene induction following HSP90 inhibition with 250 nM 17-AAG treatment. Bar graphs show the percentage induction of (A) HSPA1A mRNA or (B) HSPB1 mRNA. Percentage induction was calculated as a percentage of mean mRNA relative to the treatment with 250 nM 17-AAG and transfection with AllStars negative control siRNA condition. (C) Fold change in basal HSF1 mRNA levels following silencing of HSF1 or DHX8 (mean ± SEM, n ≥ 3) relative to negative control siRNA. (D) Immunoblot analysis of heat shock protein expression over time following siRNA-mediated knockdown of DHX8 or HSF1, following 250 nM 17-AAG treatment. GAPDH served as a loading control. AS = AllStars negative control siRNA, O1 = DHX8 siRNA 1, O3 = DHX8 siRNA 3, HSF = HSF1 siRNA. Immunoblots shown are exemplars of two independent repeats. Effects of HSF1 or DHX8 knockdown on transcript levels on non-heat shock-regulated control mRNAs: (E) NOXA, (F) MYC, and (G) CDC37 were quantified using RT-qPCR TaqMan (mean ± SD, n = 3). Impact of HSF1 or DHX8 knockdown for 72 h on RNA processing was assessed by RT-qPCR using TaqMan probes targeting intron 1 regions of (H) HSF1 or (I) NOXA (mean ± SEM, n ≥ 3, **P < .01 determined using a two-tailed t-test comparing pre-mRNA levels between AllStars negative control and DHX8 siRNA conditions. (J) Data from RIP microarray analysis showing association of HSF1 or NOXA RNA with DHX8 (mean ± SEM, n = 5 **P < .001 determined using a two-tailed t-test comparing pull-down with DHX8 antibody compared to the IgG control). Only significant differences are indicated on the plots.
Interestingly, none of the siRNA pools targeting the other 51 DEAH/D-box RNA helicases in the screening library, including canonical spliceosome-associated DHX15, DHX16, DHX38, DDX23, and DDX46 [28], strongly inhibited HSPA1A induction (Supplementary Data S1). Although pooled siRNAs against DHX16 and DHX38 showed modest effects in the primary screen (45% and 38% inhibition, respectively), retesting the four individual siRNAs revealed no activity for the four DHX38 siRNAs and only partial (~50%) inhibition by two DHX16 siRNAs (Supplementary Fig. S3A). Follow-up immunoblotting confirmed that DHX16 knockdown did not reduce basal HSF1 protein or HSP90 inhibitor-induced HSP72 expression (Supplementary Fig. S3B). Collectively, these results indicate that, among the canonical DEAH/D-box helicases, DHX8 uniquely supports HSF1 expression, or that the remaining helicases are functionally redundant.
DHX8 regulates HSF1 mRNA processing
Vertebrate DHX8 and its yeast homolog Prp22 participate in the late stages of mRNA splicing [29–35]. In the U2OS cells used for screening, DHX8 knockdown caused a marked accumulation of HSF1 pre-mRNA containing intron 1 (Fig. 2H). To confirm that this response was robust and not simply idiosyncratic to U2OS cells, we selected a second cell line, HCT116 human colorectal cancer cells. We have previously used HCT116 cells to demonstrate their dependence on different HSP70 isoforms, including the HSF1-regulated HSPA1A [11]. These cells are also described to have high HSF1 expression, which is reported to drive their malignant phenotype [36–38], and have been used in multiple phenotypic screens for HSF1 inhibitors [39–42]. Similar to the U2OS cell data, DHX8 knockdown in HCT116 cells also resulted in significant accumulation of HSF1 pre-mRNA (Supplementary Fig. S4A), supporting a role for DHX8 in HSF1 mRNA maturation. In contrast, DHX8 silencing did not lead to significant intron retention in a control transcript, NOXA mRNA in either cell line (Fig. 2I and Supplementary Fig. S4B), suggesting a selective splicing defect rather than global splicing failure.
To obtain a global overview of the interaction of DHX8 with RNAs, including HSF1 pre-mRNA, we immunoprecipitated the DHX8 protein and analysed the co-immunoprecipitated RNA. We initially performed a pilot RIP experiment using RT-PCR endpoint analysis, which revealed clear binding to HSF1 mRNA and enrichment of HSF1 mRNA containing intron 1 relative to mature HSF1 mRNA (Supplementary Fig. S5). In contrast, NOXA control mRNA showed substantially less enrichment than HSF1 in DHX8 immunoprecipitates. As a positive control, we showed HSF1 and NOXA pre-mRNAs and mature transcripts co-immunoprecipitated with snRNP70, a component of the U1 small nuclear ribonucleoprotein spliceosome complex [43] (Supplementary Fig. S5), confirming their engagement with the canonical splicing machinery. To profile RNAs that interact with DHX8, we subsequently ran the RIP assay using a microarray endpoint. Compared to the IgG control, there was a significant enrichment of HSF1 RNA (P = .0008) with the DHX8 pull-down (Fig. 2J). Together, these findings, along with our siRNA data demonstrating intron retention and HSF1 protein depletion in U2OS (Fig. 2C and D) and HCT116 cells (Supplementary Fig. S4), support a model in which DHX8 binds to the HSF1 pre-mRNA and is required for efficient splicing and maturation of HSF1 mRNA.
DHX8 is required for global HSF1-regulated gene expression
To compare global gene expression regulated by HSF1 or DHX8 with RNAs identified in the DHX8 RIP assay, we initially profiled gene expression with the same microarray platform used for the RIP assays in U2OS cells following HSF1 or DHX8 siRNA knockdown. Knockdown of HSF1 significantly altered the expression of 2101 gene probes, whereas DHX8 knockdown had a broader impact, affecting 6927 gene probes (Fig. 3A and Supplementary Data S2). Analysis of the transcription factor DNA-binding sites and transcription factor-encoding genes identified 602 transcription factor-regulated gene sets and 195 genes encoding transcription factors that were affected by DHX8 knockdown (Supplementary Data S3). Genes associated with heat shock promoter elements, but not HSF1 itself, were significantly enriched 2.7-fold (P = 1.59 e^−35^) within the subset of genes regulated by DHX8 knockdown (Supplementary Data S3). In particular, 710/1250 (57%) and 276/851 (32%) of the genes decreased or increased, respectively, by HSF1 siRNA treatment were significantly enriched (P < .00001; χ^2^) in the gene set affected by DHX8 knockdown (Fig. 3A).
DHX8 knockdown disrupts global gene expression and RNA splicing in U2OS human osteosarcoma cells. (A) Venn diagram illustrating the number of gene probes with significantly altered expression in both 48- and 72-h siRNA treatments with DHX8 or HSF1 siRNAs (Padj < .05, >1.5-fold change; Supplementary Data S2) compared to treatment AllStars negative control siRNA (determined by microarray; n = 3 independent repeats), and transcripts identified by RIP as physically associated with DHX8 compared to the mouse immunoglobulin negative control (Padj < .05, >1.5-fold change, n = 5 independent repeats; Supplementary Data S4). (B) Principal component analysis (PCA) plot depicting global splicing alterations (determined by RNA-seq) in untreated cells (mock) or treated with control (AllStars negative control), DHX8-O1 or DHX8-O3 siRNAs for 72 h. Venn diagrams comparing (C) differentially expressed genes (determined by RNA-seq; P < .01, >2-fold change; n = 3 biological repeats) or (D) individual intron retention events (Padj < .05) following treatment with DHX8-O1 or DHX8-O3 siRNAs relative to the AllStars negative control siRNA. (E) Log10 plot of P-values versus fold changes for retention of HSF1 introns 1 through 12 following treatments with DHX8-O1 siRNA. The dotted line indicates Padj ≤ .05, symbol size is proportional to fold change in intron retention relative to the untreated control (mock). Distribution plots of percent spliced in (PSI) values for splicing events showing (F) a comparison of the AllStars negative control siRNA (negative control axis) and untreated control (mock axis) or DHX8-O1 siRNA treated cells. (G) Summary of splice variant categories altered between AllStars negative control versus siRNA DHX8-O1. Grey data points are not significant while the other coloured data points indicate significantly altered splicing events (Padj < .05).
Compared to the 6927 probes affected by DHX8 silencing, fewer mRNAs (2785 probes, including HSF1 mRNA probes) were directly bound by DHX8, with 586 gene probes significantly enriched in the gene expression set affected by DHX8 knockdown (Fig. 3A; P < .00001; χ^2^). A smaller subset of mRNAs (87 probes) was bound by DHX8 in the RIP assay and modulated by both DHX8 and HSF1 knockdown and was significantly enriched in the subset of probes (986) whose expression was affected by HSF1 and DHX8 knockdown (Fig. 3A; P < .0098; χ^2^). Analysis of the genes bound by HSF1 [15], whose mRNA was identified in the DHX8–RIP assay, revealed that DHX8 preferentially binds to the mRNA of HSF1-regulated genes that are also implicated in cancer (Supplementary Data S5). Indeed, transcripts from genes in the HSF1 cancer signature were significantly overrepresented (P < .0001; χ^2^) among DHX8-bound mRNAs, suggesting that expression of this gene subset depends on DHX8.
Given the role of DHX8 in pre-mRNA splicing, we examined global splicing defects by RNA-seq profiling of U2OS cells following 72-h DHX8 knockdown. Both siRNAs induced an overlapping, high degree of altered gene expression and splicing compared to the untreated or negative controls (Fig. 3B–D). Importantly, HSF1 mRNA exhibited intron retention after treatment with both DHX8 siRNAs, with significant retention observed in most introns for at least one DHX8 siRNA, but not for the NOXA gene transcript used as a control in the RIP experiments (Fig. 3E and Supplementary Fig. S6). DHX8 knockdown and negative control siRNA conditions for the different classes of splice variant revealed that intron retention was the predominant significant splicing defect following DHX8 knockdown (Fig. 3F and G). This phenomenon has also been reported following the genetic knockout of Prp22 in yeast and, more recently, DHX8 in zebrafish [29, 44]. Additionally, we identified a significant overlap (356 mRNAs, including HSF1) between the 1312 mRNAs displaying altered splicing following DHX8 knockdown and mRNAs bound by DHX8 in the DHX8–RIP experiment (Fisher’s exact test; P < .0001, odds ratio = 2.85; Supplementary Data S6).
Gene set enrichment analysis (GSEA) revealed no significant pathway enrichment among genes with altered expression following DHX8 knockdown. In contrast, genes exhibiting altered splicing showed significant enrichment for pathways involved in vitamin D transport and signalling (reactome transport of vitamins nucleosides and related molecules—Padj = .002; Biocarta VDR pathway—Padj = .005; WP non-genomic actions of 125-dihydroxy-vitamin D3—Padj = .036). Enrichment was also observed for pathways related to protein synthesis (reactome eukaryotic translation initiation, Padj = .027; reactome nonsense mediated decay, Padj = .03; reactome selenoamino acid metabolism, Padj = .04; WP cytoplasmic ribosomal proteins, Padj = .001) and the G2/M phase of the cell cycle (hallmark mitotic spindle, Padj = .022). Finally, we examined the overlap with the CaSig1 HSF1-regulated cancer gene set (456 genes) identified by Mendillo et al. [15]. Gene expression altered by DHX8 loss showed no significant overlap (20%; P = .161; χ^2^), whereas the genes with transcripts showing aberrant splicing were significantly enriched (61%; P < .0001; χ^2^).
In summary, these data showed that, in the U2OS cell line used for screening, DHX8 loss reduced HSF1 protein and altered expression of HSF1-regulated genes. RNA-seq revealed intron retention in HSF1 pre-mRNA and HSF1-associated CaSig1 transcripts [15], along with widespread splicing defects affecting vitamin D signalling, protein synthesis, and cell-cycle pathways, indicating DHX8 functions beyond HSF1.
Mutation of DHX8 ATPase or RNA-binding sites results in DHX8 inactivation and retention of introns in HSF1 and other mRNAs
To further explore the role of DHX8 and the consequences of its loss of function on HSF1 and pre-mRNA splicing, we used the human HCT116 colorectal cancer cells described earlier. Our previous studies have shown that the K594E mutation in human DHX8 impairs ATP turnover while preserving RNA-binding, whereas hook-turn (R647G and F648G) and hook-loop (Y813G, S814G, and A815G) mutants disrupt RNA binding (Fig. 4A and B) [45]. Silencing DHX8 with siRNAs targeting either the CDS (DHX8-O1) or the 3′ untranslated region (UTR; DHX8-010) resulted in a marked loss of HSF1 protein expression (Fig. 4C). Importantly for our rescue experiments, induction of exogenous HA-tagged DHX8 protein was suppressed by the CDS-targeting siRNA but remained unaffected by the 3′UTR-targeting siRNA, since the doxycycline-inducible cDNA construct expressed mRNA that lacked the 3′UTR (Fig. 4D).
*Rescue of DHX8 knockdown: correct splicing and expression of HSF1 requires DHX8 ATPase and RNA-binding activity in HCT116 human colorectal cancer cells. (A) Structural representation of the human DHX8 helicase domain (PDB ID: 6HYT), shown as a blue ribbon. Key functional residues are highlighted: the ATPase-deficient mutation (K594E), the hook-turn RNA-binding mutants (R647G, F648G), and the hook-loop RNA-binding mutants (Y813G, S814G, A815G). The WT residues are depicted as cylinders with carbon atoms in orange, overlaid by a semi-transparent surface. An ADP molecule is shown in green, bound at the interface of the RecA1 and RecA2 domains. A poly-adenine RNA strand (light blue) from the RNA-bound DHX8 structure (PDB ID: 6HYU) is superimposed to indicate the RNA-binding tunnel. (B) Schematic of DHX8 (not to scale) showing the domains and motifs conserved in DHX8 [45]. The RS domain of DHX8 facilitates phosphorylation-dependent protein interactions that recruit and regulate DHX8 within the spliceosome. The S1 domain binds and positions pre-mRNA substrates, supporting spliceosome remodelling during splicing. Motifs I, Ia, Ib, II, and III, form the Rec1A part of the conserved helicase core and coordinate ATP binding, hydrolysis, and RNA engagement. The RecA2 domain completes the helicase core and, via motifs IV, V, and VI, stabilizes RNA-binding, links RNA engagement to ATP-driven conformational shifts, and detects nucleotide state to enable efficient, directional RNA translocation. Additional domains, such as the winged-helix, HA2, and OB-fold, provide auxiliary RNA-binding surfaces and regulatory scaffolding that stabilize RNA–RNP interactions, fine-tuning helicase activity during RNA remodelling. The positions of the mutations used in this study are indicated. (C) Validation of DHX8 siRNAs designed to target the CDS (DHX8-O1; Supplementary Table S2), which affects both endogenous and exogenously expressed DHX8, or the 3′UTR (DHX8-O10; Supplementary Table S2), which is unique to endogenous DHX8 (the immunoblot is an exemplar of n = 3 independent repeats). (D) Capillary immunoassay analysis of endogenous DHX8, exogenous doxycycline-inducible HA-tagged DHX8, HSF1, and GAPDH (loading control). With this immunoassay platform, protein expression levels are directly comparable between different clones and experimental conditions as all samples are loaded and analysed from a single multiwell plate with repeated sampling for each antigen analysis. HCT116 cells transduced with a lentivirus encoding doxycycline-inducible WT DHX8, the ATPase-deficient K594E mutant, or RNA-binding mutants (hook turn: R647G_F648G; hook loop: Y813G_S814G_A815G) were analysed following treatment with negative control, UTR or CDS siRNAs ± 1 μg/ml doxycycline (Dox). The immunoassay readout is an exemplar of n = 2 independent repeats. (E) Cells were treated with either scramble negative control (neg. con.), CDS, or 3′UTR siRNAs without (–Dox) or with (+Dox) doxycycline induction. RT-qPCR TaqMan analysis of total HSF1 mRNA and intron-retained pre-mRNA in the WT, ATPase K594E, and hook turn (R647G, F648G) RNA-binding mutants. HCT116 cells were treated with the 3′UTR siRNA in the presence or absence of doxycycline induction (mean ± SD, n = 4). Statistical significance was assessed using an unpaired two-tailed Mann–Whitney test (P < .05; only significant differences are indicated). (F) Integrative Genomics Viewer (IGV) plot of HSF1 transcript structure following RNA-seq analysis in HCT116 cells treated with either scramble negative control (blue) or 3′UTR siRNA (red).
Induction of WT DHX8 successfully restored HSF1 protein levels and reduced the accumulation of intron-retained HSF1 pre-mRNA following endogenous DHX8 depletion (Fig. 4D and E). In contrast, induction of ATPase-deficient or RNA-binding DHX8 mutants failed to rescue HSF1 protein expression or pre-mRNA splicing. Moreover, the hook-turn RNA-binding mutant caused a significant increase in the intron-retained HSF1 pre-mRNA accumulation (Fig. 4D and E). There were no major differences in the induction of the HA-tagged DHX8 protein, indicating that poor induction of the rescue protein did not account for the failure to rescue HSF1 biomarker changes following knockdown of the endogenous DHX8 (Fig. 4D).
RNA-seq analysis of HCT116 cells treated with DHX8 siRNA without rescue found significantly altered gene expression (Supplementary Fig. S7) and splicing, although only a small subset of genes showed simultaneously altered expression and splicing (Fig. 5A and B). Consistent with the findings in U2OS osteosarcoma cells (Fig. 3E–G), the predominant splicing defect in HCT116 colorectal cells was intron retention (Fig. 5B). Reintroduction of WT DHX8 rescued both gene expression and splicing defects, reducing the number of genes showing altered expression or splice junctions with disrupted splicing following knockdown of endogenous DHX8 (Fig. 5A–C and Supplementary Fig. S7). However, the ATPase-dead K549E mutant or the RNA-binding-defective R647G_F648G mutant both failed to rescue the DHX8 knockdown effects and instead exacerbated splicing defects (Fig. 5B and C), particularly intron retention, as previously observed for HSF1 (Fig. 4E). These findings highlight the essential roles of both the RNA-binding and ATPase activities of DHX8 in proper splicing and expression of HSF1 and a subset of the cellular transcriptome.
DHX8 binds pre-mRNAs between the lariat branch site and 3′ splice junction and requires ATPase activity and RNA-binding activities for accurate intron removal from pre-mRNAs. (A) PCA of RNA-seq global gene expression profiling of total RNA from HCT116 human colorectal cancer cells (Fig. 4D and E) following treatment with the 3′UTR siRNA (DHX8-O10; n = 4 independent repeats; each symbol corresponds to an individual sample). (B) RNA-seq quantification of global changes in gene expression (P < .05, >2-fold change) or differential splicing events (delta PSI > 10%, >0.95 confidence) upon DHX8 knockdown and rescue following induction of WT, ATPase-dead, or RNA-binding mutant DHX8. For each WT and mutant rescue model, data are plotted relative to the negative control siRNA (Supplementary Data S7–S14). (C) Plot highlighting DHX8-dependent splicing events disrupted by DHX8 knockdown and rescued by re-expression of WT or mutant DHX8. (D) Meta-intron profile showing the distribution of DHX8 binding across pre-mRNAs in HCT116 cells, identified by eCLIP (data show 555 of the 1138 significant peaks; P < .05, three-fold increase from the n = 3 independent repeats). The inset in the metaintron plot shows top-ranked conserved RNA sequence motifs significantly enriched at DHX8-binding sites (1097 target sequences versus 37 080 background sequences, P < 1e–10). (E) Distribution of DHX8-binding sites on pre-mRNAs/mRNAs identified by eCLIP (n = number of peaks associated with each feature; log2-fold enrichment ≥3, P ≤ .001; Supplementary Data S15), highlighting preferential binding to intronic regions near splice junctions. Venn diagrams illustrating the overlap of DHX8-bound transcripts identified by eCLIP and genes displaying (F) differential gene expression or (G) all alternative splicing events following DHX8 knockdown (P < .0001, odds ratio = 3.7, Jaccard Index = 0.1).
DHX8 binds select pre-mRNAs between the lariat branch and 3′ splice junction
We applied eCLIP [46] to profile DHX8 RNA binding in HCT116 cells and identified 716 significant RNA-binding sites repeatedly enriched at 5′ splice junctions and within protein-coding regions (Fig. 5D and E, and Supplementary Fig. S8). Comparison to the RIP data at the gene level found 157 (22%) RNAs were significantly enriched in the eCLIP and RIP experiments (P < .00001; χ^2^). Motif analysis revealed conserved sequences, including 5′ and 3′ splice junction motifs, as well as a polypyrimidine tract typically found between the lariat branch site and the 3′ splice junction (Fig. 5D and E) [28]. DHX8 binding was predominantly localized to the intronic side of the 5′ splice junctions, with minimal binding on the exonic side (Supplementary Fig. S9A). In contrast, the binding at the 3′ splice sites were evenly distributed across the junction. This pattern suggests that DHX8 interacts with lariat-containing pre-mRNA, likely between the intronic lariat branch point and the 3′ junction (Supplementary Fig. S9B), supporting its role in the second transesterification step of splicing and explaining the increased intron retention observed following DHX8 knockdown in U2OS (Fig. 3F and G) and HCT116 cells (Fig. 5B).
eCLIP is highly effective for mapping RNA–protein interactions, but it has limitations, including variability in immunoprecipitation, crosslinking, RNA-binding affinity, and target RNA abundance [46]. Although we detected significant binding of DHX8 to HSF1 intronic sequences in two of the three repeats, the binding peaks were insufficiently enriched across all three samples to be scored as significant in the final analysis of all three independent repeats (Supplementary Fig. S10). However, we found significant enrichment (~ 37%; P = 4e^−52^, OR = 3.7, Jaccard index = 0.1) of gene transcripts with DHX8 binding among those exhibiting altered splicing, but not among genes showing altered expression (Fig. 5F and G). Together, the RNA-seq and eCLIP data show that DHX8 helicase activity and RNA binding between the lariat and 3′ splice junction are required for correct splicing of a subset of transcripts, including HSF1.
Targeting DHX8 mRNA by siRNA, or DHX8 protein by dTAG degradation, results in similarly altered gene expression and disrupted splicing
Depletion of DHX8 using siRNA caused pronounced intron retention, but the overlap between transcripts with splice-junction changes and altered expression was limited. This disconnect may reflect technical constraints in short-read RNA-seq splice analysis and/or secondary effects downstream of the loss of DHX8 that accumulate during the 72-h siRNA treatment. Therefore, we decided to use an orthogonal DHX8–dTAG degron-based strategy, potentially enabling rapid and near-complete target protein depletion to define the role of DHX8 while potentially minimizing secondary effects.
We selected human H1299 NSCLC cells as an additional model, as they have been used to study HSF1 activation and show sensitivity to DHX8 siRNA knockdown, with robust induction of HSF1 intron retention and reduced cell growth (Fig. 6A and Supplementary Fig. S11) [47, 48]. As described earlier, no other canonical spliceosome-associated DEAD/H-box helicase was identified as a hit in the U2OS screen (Supplementary Fig. S3 and Supplementary Data S1). Similarly, in H1299 cells, we demonstrated that siRNA targeting the spliceosome-associated DEAD/H-box helicases (DDX23, DDX46, DHX15, DHX16, and DHX38) failed to induce HSF1 intron retention to the same degree as DHX8 knockdown and generally did not influence each other’s expression (Supplementary Fig. S11).
Acute DHX8 degradation in H1299 human NSCLC cells at 6-h recapitulates the effects of long-term DHX8 siRNA knockdown at 72 h. (A) Real-time proliferation of H1299 human NSCLC cells treated with DHX8, positive death-inducing control, or negative control siRNA (determined using Incucyte). Data represent mean ± SEM from three biological replicates. (B) Expression of the HiBit tagged DHX8–dTAG protein following continuous exposure to a range of dTAGV-1 concentrations for 1–7 days (n = 2 biological repeats). The tagged protein was detected by HiBit luminescence. (C) Real-time proliferation of H1299 DHX8–dTAG model following continuous exposure to a range of dTAGV-1 concentrations. Growth was monitored over time and is shown as mean ± SEM from three biological replicates. (D) Immunoblot of DHX8–dTAG, HSF1, and HSP72 following treatment with vehicle control (–) or 250 nM dTAGV-1 (+) for 8, 16, and 24 h. Actin is included as a loading control. The immunoblot is an exemplar of two independent repeats. (E) IGV tracks displaying HSF1 splicing profiles in H1299 cells treated with DMSO (blue) or dTAGV-1 (red) for 6 h, based on RNA-seq analysis. Venn diagrams illustrating the overlap of (F) differentially expressed genes (P < .05, >2-fold change; P < .0001, odds ratio = 1.2, Jaccard Index = 0.2) and (G) alternative splicing events (delta PSI > 10%, >0.95 confidence; P < .0001, odds ratio = 9.5, Jaccard Index = 0.3) in H1299 cells treated with DHX8 siTOOLs pooled siRNA for 72 h or dTAGV-1 for 6 h, relative to respective controls (negative control siRNA or DMSO vehicle control, respectively; n = 4 independent repeats; Supplementary Data S16–S19). (H) Bar graphs summarizing the categories and frequencies of transcriptional and splicing changes observed across treatments, based on data from panels (F) and (G). (I) Venn diagram depicting the overlap in splicing events detected after DHX8 knockdown in HCT116 and H1299 cells (72-h siRNA), short-term DHX8 degradation in H1299 cells (6-h dTAGV-1), and DHX8-bound transcripts identified by eCLIP in HCT116 colorectal cells.
After confirming that HSF1 transcript processing in H1299 cells depends on DHX8, as in U2OS and HCT116 cells, we generated H1299 single-cell clones in which endogenous DHX8 was replaced with an exogenous DHX8–dTAG fusion protein. This fusion protein was tagged at the N-terminus with the mutant FKBP12^F36V^ and the HiBiT small peptide, enabling rapid detection of fusion protein expression. Treatment of either clone with the small-molecule heterobifunctional degrader dTAG^V^-1 [49] resulted in a concentration-dependent loss of the fusion protein and inhibition of cell growth out to 7 days (Fig. 6B and C, and Supplementary Fig. S12A–C). Importantly, in both clones, loss of the DHX8–protein following 24 h of treatment with dTAG^V^-1 led to decreased expression of HSF1 and HSP70 (Supplementary Fig. S12D); this was a time-dependent effect with a limited decrease detected 8 h post-treatment that decreases further at 16 and 24 h (Fig. 6D).
To compare sustained and acute DHX8 loss, we profiled gene expression and splicing 72 h after DHX8 siRNA knockdown or 6 h after dTAG^V^-1 treatment of DHX8–dTAG (Supplementary Fig. S12E). Similar to the increased intron retention following siRNA knockdown of DHX8 (Supplementary Fig. S11), acute dTAG degradation of DHX8 at 6 h also induced intron retention in the HSF1 transcript (Fig. 6E). Both perturbations caused broadly significantly similar altered gene expression (P < .0001; χ^2^), with ~1/3 of the differentially expressed genes being shared (Fig. 6F). Interestingly, acute DHX8 degradation by dTAG^V^-1 induced ~30% more alternative splicing events, of which ~1/3 significantly overlapped with the siRNA-induced events (P < .0001; χ^2^; Fig. 6G). As before, with both siRNA- and dTAG-induced loss of DHX8 there was limited overlap between genes showing altered expression and those showing altered splicing (Fig. 6H). Intron retention was the most prominent splicing defect in both types of DHX8-loss, including disrupted splicing of the HSF1 transcript (Fig. 6E and H). Integrating DHX8 eCLIP data from HCT116 cells revealed that ~43% of the DHX8-bound transcripts exhibited splicing alterations under at least one perturbation, and 159 (22%) of the DHX8-bound transcripts identified by eCLIP showed disrupted splicing under all conditions (P < .0001 χ^2^; Fig. 6I).
GSEA of datasets from the HCT116 and H1299 cell-profiling experiments revealed distinct impacts of DHX8 loss on gene expression and splicing profiles. The sets of significantly enriched pathways also showed limited overlap between gene expression and splicing data (Supplementary Fig. S13A and B). In the gene expression data, increased expression of NF-κB target genes in response to TNF-α was the only gene set to show highly significant enrichment across multiple DHX8 perturbation conditions (Supplementary Fig. S13A). In contrast, aberrant splicing was consistently and highly significantly associated with gene sets related to the unfolded protein response, DNA repair, the mitotic spindle, the G2M checkpoint, and E2F/cMYC targets (Supplementary Fig. S13B). DHX8 loss also perturbed splicing of p53-pathway transcripts in the TP53 WT HCT116 [50], but not in the TP53-null H1299 cells [51]. GSEA of genes with both significantly altered expression and aberrant splicing identified few enriched pathways, observed only in the H1299 dTAG and HCT116 mutant rescue models (Supplementary Fig. S13C). In the dTAG model, G2M checkpoint, cMYC targets, and E2F targets were the most significantly enriched gene sets, whereas the p53 pathway was enriched in the HCT116 rescue model; these pathways were also significantly altered in the splicing GSEA analysis (Supplementary Fig. S13B and C).
Finally, we also examined expression and splicing of the 456 CaSig1 HSF1-regulated, cancer-associated genes linked to poor prognosis [15]. In the expression dataset, 130 genes (29%) showed altered expression in at least one condition, with 11 altered across all conditions (not significant, P = .159; χ^2^; Supplementary Fig. S13D). In contrast, overlap with transcripts exhibiting impaired splicing was greater, affecting 222 genes (49%) in at least one condition, of which 50 (11%) were significantly altered across all conditions (P < .0001; χ^2^; Supplementary Fig. S13E).
Overall, these data show that multiple orthogonal approaches disrupting DHX8 expression or function predominantly cause defective splicing via intron retention in transcripts encoding HSF1, HSF1-regulated cancer genes, and genes associated with oncogene activation and stress.
DHX8 knockdown inhibits cancer cell proliferation and induces apoptosis in cancer cell lines
In the previous section, we showed that loss of DHX8, either by siRNA knockdown (Fig. 6A) or dTAG-degradation (Fig. 6C and Supplementary Fig. S12C), inhibited cell proliferation. Silencing DHX8 for 72 h caused an accumulation of U2OS cells in the G2/M phase of the cell cycle (Fig. 7A and Supplementary Fig. S14). Additionally, increased levels of cleaved PARP and caspase 3, apoptotic markers, were detected in DHX8-depleted U2OS cells (Fig. 7B).
*DHX8 knockdown impairs HSF1-dependent heat shock protein induction and proliferation and survival of cancer cell lines. (A) Cell cycle distribution of U2OS human osteosarcoma cells 72 h post-transfection with DHX8 or AllStars control siRNA. Bar graph quantifies the proportion of cells in G1, S, and G2/M phases, revealing G2/M phase accumulation following DHX8 knockdown (mean ± SD, n = 3). (B) Immunoblot showing increased levels of cleaved PARP and caspase-3 following DHX8 knockdown, indicative of apoptosis. (C) Immunoblot analysis of HSF. 1, HSP72, and HSP27 protein levels before and after treatment with 250 nM HSP90-inhibitor 17-AAG to activate HSF1, following DHX8 knockdown. Tumorigenic: U2OS (osteosarcoma), MDA-MB-231 (breast), and HCT116 (colorectal). Non-tumorigenic: MCF10A (human mammary epithelial) and CCD-18Co (human colon fibroblast). The immunoblots shown in panels (B) and (C) are representative of two independent repeats, with GAPDH as a loading control. (D) Concentration-responsiveness of HSPA1A mRNA quantified by TaqMan qPCR following 6-h treatment with 17-AAG (mean ± SEM, n = 3; dotted line indicates 250 nM 17-AAG used in HSF1-induction experiments). (E) Induction of HSPA1A mRNA by 250 nM 17-AAG (6-h treatment) across tumorigenic and non-tumorigenic cell lines. Induction was calculated as a percentage of mean mRNA relative to treatment with the 250 nM 17-AAG and transfection with the AllStars negative control siRNA (indicated by the dotted black line). (F) Cell viability measured by CellTiter-Glo assay following siRNA-mediated knockdown of DHX8 or (G) HSF1 for 3 or 7 days. Apoptosis measured by Caspase 3/7-Glo assay following knockdown of (H) DHX8 or (I) HSF1 for 3 or 7 days. (F–I) Data are plotted relative to AllStars negative control siRNA treatment condition (indicated by the dotted black line). Bar graphs display mean ± SEM (n ≥ 3 independent repeats). Statistical significance was determined using a two-tailed t-test comparing paired tumorigenic and non-tumorigenic cell lines (ns = not significant; *P < .05; **P < .01; **P < .001). Only significant differences are indicated on the plots.
To further examine the effect of DHX8 knockdown on heat shock protein induction, proliferation, and survival, we analysed cancerous and non-immortalized human cell lines. In a mini-panel of human osteosarcoma, colorectal and breast cancer cell lines and non-tumorigenic lines, we found HSF1 protein levels were significantly reduced by DHX8 knockdown in all three cancer cell lines (U2OS osteosarcoma, HCT116 colorectal, and MDA-MB-231 breast) but were less affected in non-tumorigenic lines (MCF10A breast and CCD-18Co colon; Fig. 7C). Furthermore, induction of HSP72 and HSP27 following HSP90 inhibition was more strongly inhibited by DHX8 knockdown in tumorigenic cells than in non-tumorigenic cells (Fig. 7C).
Despite similar basal HSPA1A mRNA levels (Supplementary Fig. S15A), 17-AAG-mediated HSPA1A mRNA induction was lower in the non-tumorigenic lines (Fig. 7D), suggesting that HSF1 is more readily activated in tumour cells. Silencing HSF1 reduced HSF1 mRNA levels equally across all cell lines (Supplementary Fig. S15B) and decreased HSF1-induced HSPA1A expression in both tumour and non-tumorigenic cells but had a greater effect in the tumour cell lines (Fig. 7E). DHX8 knockdown also suppressed HSF1-mediated induced expression of HSPA1A in cancer cell lines but had minimal effect in the non-tumorigenic lines (Fig. 7E).
We then investigated whether silencing of DHX8 or HSF1 differentially affected proliferation and viability. DHX8 knockdown had (i) a more pronounced anti-proliferative effect in tumorigenic cell lines than in non-tumorigenic lines (Fig. 7F), and (ii) a greater impact on proliferation than HSF1 knockdown (Fig. 7F and G). Apoptosis was induced in U2OS osteosarcoma and MDA-MB-231 breast cancer cell lines, but not in HCT116 colorectal cancer cells or in either non-tumorigenic line (Fig. 7H). In contrast, HSF1 knockdown did not induce significant apoptosis in any of the five cell lines analysed (Fig. 7I).
Overall, these findings indicate that targeting DHX8 can inhibit the proliferation and survival of cancer cells and highlight DHX8 as a potential therapeutic target for selective inhibition of cancer cell proliferation and survival.
Discussion
HSF1 is a promising target in oncology; however, as a ligandless transcription factor, it has been challenging to inhibit pharmacologically [6, 7, 17]. To identify potentially druggable regulators of HSF1-dependent stress responses in cancer cells, we conducted an siRNA screen for modulators of HSF1-mediated HSPA1A expression. A pre-mRNA splicing factor, the DEAH-box RNA helicase (DHX8), was the most robustly validated hit, and significantly modulated HSF1 activity and reduced the expression of HSF1-dependent genes. This identified DHX8 as a previously unrecognized regulator of HSF1 and a potential therapeutic target in this pathway.
To date, only a few studies have linked splicing factors to HSF1 regulation, primarily using siRNA screens in Caenorhabditis elegans and human HEK-293 and HeLa cells, where SF3B1, a key component of the spliceosome, was identified as a regulator of HSF1 activation following heat shock [52–55]. Pladienolide B, an SF3B1 inhibitor produced by bacteria, suppresses HSF1 protein [54, 56]. Consistent with this, we also found that SF3B1 knockdown caused intron retention in the HSF1 pre-mRNA, similar to DHX8 knockdown.
Following our identification and validation of DHX8 as a modulator of HSPA1A expression upon HSF1 activation, we demonstrated that silencing, mutation, or degradation of DHX8 across multiple human cancer cell lines led to intron retention in HSF1 pre-mRNA and a concomitant decrease in HSF1 protein levels. Furthermore, a RIP profiling assay confirmed the binding of DHX8 to the HSF1 pre-mRNA, supporting a direct role in its splicing. Rescue experiments using DHX8 mutants deficient in RNA-binding or ATPase activity showed that these DHX8 activities are essential for restoring HSF1 splicing.
DHX8 was not identified in earlier siRNA screens targeting heat shock response pathways [52–55]. Our siRNA library focused on proteins predicted to be druggable. Although it did not include SF3B1, it included 51 DEAD/DEAH-box RNA helicases. Among these, only DHX8 emerged as a significant hit, while further validation in a second cell line confirmed that knockdown of other RNA helicases with known function in the spliceosome did not affect HSF1 pre-mRNA splicing. Therefore, our findings indicate that HSF1 pre-mRNA splicing has a unique dependence on DHX8 compared with other canonical splicing helicases.
Cryo-EM studies have identified sequential spliceosome states: pre-B, B, Bact, B*, C, C*, P, and ILS [28, 57–63]. Splicing involves two catalytic steps: In the first step, the B* complex catalyses the cleavage of the 5′ exon and the formation of a lariat–3′ exon intermediate. This generates the C complex that undergoes remodelling to form the C* complex, which catalyses the second step, producing ligated exons and a spliced-out intron. In the C* complex, the yeast homologue of DHX8 (Prp22), binds to the lariat intermediate in the intron upstream of 3′-splice junction that may reflect dynamic sampling and proofreading of 3′ splice sites through repeated Prp22 binding and dissociation [64]. Subsequently, Prp22 binding shifts to a narrower region of 10–25 nucleotides of the exon downstream of the 3′-splice site, where it may pull on the 3′-exon to release mRNA from the spliceosome [65]. Early cryo-EM studies located DHX8 on the periphery of the human spliceosome C* complex, while more recent analysis identified two intermediate complexes, pre-C*-I and pre-C*-II, where DHX8 occupies distinct positions across these complexes and the C* complex [60, 61]. These findings are consistent with roles for DHX8 in 3′ exon binding, proofreading, and release of mRNA from the spliceosome [61, 66, 67]. This aligns with our eCLIP data showing DHX8 binding across a region encompassing the lariat branch point and the 3′ splice junction, which also closely resembles the binding pattern observed for Prp22. Collectively, our eCLIP data and orthogonal DHX8 perturbation experiments across multiple cell lines support a model in which DHX8 loss or inhibition, in addition to impairing mRNA release from the spliceosome, disrupts 3′ exon junction proofreading and leads to incomplete pre-mRNA splicing resulting in intron retention [65].
In addition to affecting HSF1 expression, DHX8 loss was found to significantly alter HSF1-regulated gene expression and the splicing of the CaSig1 cancer-associated gene set [15]. This HSF1-regulated signature is significantly associated with poorer outcomes across multiple tumour types, implicating HSF1 in the regulation of core processes fundamental to tumour biology and prognosis. These observations are supported by studies demonstrating that HSF1 is required for oncogenic transformation in genetically engineered mouse models, leading to the concept that HSF1 drives a cancer-specific transcriptional program essential for overcoming the proteotoxic and metabolic stresses associated with oncogene activation [14, 15]. Together, these findings suggest that targeting DHX8 could represent a strategy to indirectly target HSF1 activity and its role in enabling tumour cells to cope with oncogene-induced stress. However, given the established role of DHX8 in the spliceosome, it is unlikely that regulation of HSF1 activity and output is its sole function. Indeed, our global RNA profiling, conducted following DHX8 perturbation using long-term siRNA or short-term dTAG approaches, revealed widespread altered splicing of ~2000–3000 pre-mRNAs following DHX8 loss. These included genes regulated by the cMYC oncogene and gene sets associated with the cell cycle, stress, or damage-responses. These findings align with our observations of cell accumulation in the G2/M phase following DHX8 knockdown, and also previous reports of defects in cell division and mitosis in zebrafish carrying DHX8 mutations, as well as following the knockdown of human DHX8 [29, 32, 68].
DHX8 perturbation had a more pronounced impact on cancer cells than on their non-transformed counterparts. Similarly, two independent siRNA functional screening studies in human esophagogastric and NSCLC models have also reported selective effects of DHX8 knockdown on viability in tumour versus normal cell line controls [69, 70]. In contrast, direct knockdown of HSF1 produced only relatively modest effects on cancer cell viability. These findings suggest that, at least in vitro, loss of HSF1 alone following DHX8 depletion cannot fully account for the heightened sensitivity of tumorigenic cells to DHX8 loss. This aligns with our broader observation that DHX8 depletion disrupts additional pathways beyond HSF1, which likely contributes to the cancer cell-selective vulnerability. Consistent with this finding, DHX8 was identified in an RNAi screen for synthetic lethality with mutant KRAS, suggesting that DHX8 may collaborate with oncogenic proteins to sustain cancer cell survival [71].
Our rescue experiments clearly demonstrated that both RNA binding and ATPase activities of DHX8 are essential for maintaining HSF1 protein expression. Based on this, we propose that targeting ATPase activity or ATPase-dependent helicase activity of DHX8 will have a therapeutic benefit in tumours with heightened stress responses or where aberrant splicing contributes to oncogenesis or therapy resistance. Inhibiting DHX8 to reduce HSF1 activity could be particularly effective in cancers where HSF1 is overexpressed or is critical for tumour cell transformation, proliferation, or survival [3, 7, 15, 17, 72, 73]. Furthermore, as alternative splicing of oncoproteins contributes to multiple cancer phenotypes, understanding the roles of different RNA processing patterns, pathways, and proteins in cancer is an area of considerable relevance to therapeutic intervention, and identifying which oncoproteins are regulated by DHX8 represents a promising direction for therapeutic development [74–76].
This study is limited by its reliance on a small set of in vitro models and on traditional short-read RNA sequencing, which involves computational assembly of transcripts, limiting the accurate reconstruction of full-length transcripts and, consequently, potentially missing essential features of isoform transcript architecture and expression. These limitations can be addressed in future research by employing larger panels of cancer cell lines, normal tissue and cancer organoid systems, and in vivo models, such as human tumour xenografts of the DHX8–dTAG degradation platform in mice, alongside long-read sequencing technologies that enable direct, high-accuracy analysis of full-length, alternatively spliced RNA isoforms. Also, our screen was restricted to an siRNA library targeting proteins that, at the time of its design, were considered druggable. Since then, advances in strategies for targeting protein–protein interactions, along with the development of protein glues and targeted degraders, have expanded the range of tractable targets. As a result, a whole-genome screen may have identified additional candidates for future drug discovery.
In addition to its role in canonical splicing, Sakai and colleagues recently reported that DHX8 has functions beyond splicing, linking DHX8 to cellular stress responses, nucleotide recycling, and the prevention of aberrant RNA accumulation that could disrupt cellular function [77]. The ATPase-dependent helicase activity of DHX8 is necessary for remodelling structured or abnormal RNA substrates and facilitating their delivery to the lysosome for degradation via RNautophagy [77]. This activity underscores the vital role of DHX8 in maintaining RNA homeostasis by ensuring efficient RNA turnover and preventing the accumulation of dysfunctional RNAs.
Against this backdrop, our findings reveal a previously unrecognized and critical role for DHX8 in regulating HSF1 expression in cancer cells. These data expand the biological importance of DHX8 in cancer cells and, as part of this broader function, establish a direct link between RNA processing and the transcriptional stress responses and cancer-associated gene signatures governed by HSF1. We anticipate that these findings will stimulate further investigation into the broader functions of DHX8 and the mechanistic interplay between RNA metabolism and cellular homeostasis, proteostasis, and HSF1-regulated oncogenic stress responses. More broadly, our results imply that widespread roles in mRNA processing may underlie cancer cell dependence on DHX8. Although normal cell toxicity is a potential concern when targeting a spliceosome-associated protein such as DHX8, we found that the majority of pre-mRNAs retain the capacity for successful splicing in the absence of DHX8. These observations suggest that future efforts should focus on fully characterizing the roles of DHX8 in both cancerous and normal contexts, as well as the discovery of selective chemical tools to inhibit DHX8. Collectively, such advances may ultimately pave the way for new therapeutic strategies that exploit cancer-specific vulnerabilities in RNA processing.
Supplementary Material
zcag008_Supplemental_Files
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1Richter K, Haslbeck M, Buchner J. The heat shock response: life on the verge of death. Mol Cell. 2010;40:253–66. 10.1016/j.molcel.2010.10.006.20965420 · doi ↗ · pubmed ↗
- 2Luo J, Solimini NL, Elledge SJ. Principles of cancer therapy: oncogene and non-oncogene addiction. Cell. 2009;136:823–37. 10.1016/j.cell.2009.02.024.19269363 PMC 2894612 · doi ↗ · pubmed ↗
- 3Fiser O, Muller P. Role of HSF 1 in cell division, tumorigenesis and therapy: a literature review. Cell Div. 2025;20:11. 10.1186/s 13008-025-00153-1.40287736 PMC 12034185 · doi ↗ · pubmed ↗
- 4Dai C, Sampson SB. HSF 1: guardian of proteostasis in cancer. Trends Cell Biol. 2016;26:17–28. 10.1016/j.tcb.2015.10.011.26597576 PMC 4722819 · doi ↗ · pubmed ↗
- 5Nagel R, Semenova EA, Berns A. Drugging the addict: non-oncogene addiction as a target for cancer therapy. EMBO Rep. 2016;17:1516–31. 10.15252/embr.201643030.27702988 PMC 5090709 · doi ↗ · pubmed ↗
- 6Pasqua AE, Sharp SY, Chessum NEA et al. HSF 1 Pathway inhibitor clinical candidate (CCT 361814/NXP 800) developed from a phenotypic screen as a potential treatment for refractory ovarian cancer and other malignancies. J Med Chem. 2023;66:5907–36. 10.1021/acs.jmedchem.3c 00156.37017629 PMC 10150365 · doi ↗ · pubmed ↗
- 7Workman P . Reflections and outlook on targeting HSP 90, HSP 70 and HSF 1 in cancer: a personal perspective. Adv Exp Med Biol. 2020;1243:163–79.32297218 10.1007/978-3-030-40204-4_11 · doi ↗ · pubmed ↗
- 8Dai C . The heat-shock, or HSF 1-mediated proteotoxic stress, response in cancer: from proteomic stability to oncogenesis. Philos Trans R Soc Lond B Biol Sci. 2018;373:20160525.29203710 10.1098/rstb.2016.0525 PMC 5717525 · doi ↗ · pubmed ↗