XRN1 deletion induces PKR-dependent cell lethality in interferon-activated cancer cells

SUMMARY

Emerging data suggest that induction of viral mimicry responses through activation of double-stranded RNA (dsRNA) sensors in cancer cells is a promising therapeutic strategy. One approach to induce viral mimicry is to target molecular regulators of dsRNA sensing pathways. Here, we show that the exoribonuclease XRN1 is a negative regulator of the dsRNA sensor protein kinase R (PKR) in cancer cells with high interferon-stimulated gene expression. XRN1 deletion causes PKR pathway activation and consequent cancer cell lethality. Disruption of interferon signaling with the JAK1/2 inhibitor ruxolitinib can decrease cellular PKR levels and rescue sensitivity to XRN1 deletion. Conversely, interferon-β stimulation can increase PKR levels and induce sensitivity to XRN1 inactivation. Lastly, XRN1 deletion causes accumulation of endogenous complementary sense/anti-sense RNAs, which may represent candidate PKR ligands. Our data demonstrate how XRN1 regulates PKR and how this interaction creates a vulnerability in cancer cells with an activated interferon cell state.

Graphical abstract

In brief

Zou et al. identify a requirement for the XRN1 exoribonuclease in human cancer cell lines with high expression of interferon-stimulated genes. XRN1 depletion causes accumulation of complementary sense/anti-sense transcript pairs and cell-intrinsic activation of the interferon-inducible double-stranded RNA sensor protein kinase R, leading to cancer cell lethality.

INTRODUCTION

Anti-tumor immune responses rely on the ability of the immune system to recognize tumor cells as non-self.¹ For example, adaptive immune cells can recognize cancer neoantigens as non-self to initiate tumor cell killing.² This mechanism of self/non-self recognition underlies the efficacy of immune checkpoint inhibitors and adoptive cellular therapies, which have improved clinical outcomes for patients with hematologic and solid tumor malignancies.^1,2 Similar to the adaptive immune system, the innate immune system also possesses mechanisms to distinguish self from non-self,³ but it is unclear whether these pathways can be leveraged for cancer therapeutics.

Recent studies have suggested that specific genetic or pharmacologic perturbations can induce the expression of endogenous immunogenic RNAs that can activate innate immune double-stranded RNA (dsRNA) sensing pathways in cancer cells,^4–15 a process termed “viral mimicry.”^4,14,15 Engagement of these RNA sensing pathways can, in turn, cause direct cancer cell cytotoxicity and/or stimulate anti-tumor adaptive immune responses. Disruption of epigenetic regulation,^4–6,10 perturbation of RNA splicing,¹¹ and inhibition of protein arginine methylation¹³ can all trigger the expression of dsRNA, including endogenous retroelements, and stimulate dsRNA sensing and interferon responses to inhibit cancer cell growth.

In parallel with these studies, our group and others have shown that depletion of the adenosine deaminase acting on RNA (ADAR1) causes activation of the dsRNA sensor protein kinase R (PKR) in cancer cell lines with high levels of interferon-stimulated gene (ISG) expression.^7,8,12 PKR is encoded by the EIF2AK2 gene and normally detects dsRNA of viral origin. Binding of viral dsRNA ligands triggers PKR dimerization and autophosphorylation, activating its kinase function to initiate signals that ultimately inhibit cellular protein translation, thereby restricting viral replication.^16,17 Because PKR detects dsRNA in a sequence-agnostic manner, endogenous cellular dsRNA can also stimulate its activation.^16,18 With regard to this point, a subsequent study showed that DNA hypomethylating agents can induce a requirement for ADAR1 in cancer cells through the induction of inverted Alu elements,¹⁹ which, together with mitochondrial RNA,²⁰ are known activating ligands of PKR. However, the range of cellular processes that regulate PKR activation and the identity of endogenous RNAs capable of stimulating PKR remain incompletely defined.

We hypothesized that cancer cell lines displaying a requirement for ADAR1 may also require other molecular regulators of dsRNA sensors for their survival. In this study, we analyzed genome-scale CRISPR-Cas9 gene essentiality screening data across hundreds of cancer cell lines and identified a correlation between the requirement for ADAR1 and the XRN1 exoribonuclease in cancer cell lines with high levels of ISG expression. Mechanistically, XRN1 deletion activates PKR, which was required for cancer cell lethality after XRN1 depletion. To understand the role of interferon pathway activation, we showed that inhibition of JAK1/2 signaling can decrease cellular PKR levels and rescue XRN1 knockout (KO) sensitivity. In conjunction, stimulation of interferon signaling in XRN1 KO-insensitive cell lines increased PKR levels and induced sensitivity to XRN1 deletion in a PKR-dependent manner. Computational analysis of RNA sequences revealed an accumulation of complementary sense/anti-sense RNA transcripts after XRN1 deletion. We propose that XRN1 deletion in cancer cell lines with high levels of the interferon-stimulated dsRNA sensor PKR induces accumulation of complementary sense/anti-sense RNA transcripts, thereby causing PKR activation and cancer cell lethality.

RESULTS

The enzymatic activity of XRN1 is required for survival of cancer cell lines with high expression of ISGs

To uncover molecular regulators of dsRNA sensors that may function similarly to ADAR1, we analyzed genome-scale CRISPR-Cas9 screening data generated by the Cancer Dependency Map.²¹ XRN1 genetic dependency was the top correlate with ADAR1 genetic dependency across cancer cell lines, followed by PRKRA and USP18 (Figure 1A). Notably, this correlation between ADAR1 and XRN1 dependency held across multiple cancer lineages and degrees of gene essentiality, with an R² of ~0.48 (Figure 1B). XRN1 encodes an exoribonuclease that functions as the major pathway for cytoplasmic 5′ to 3′ RNA decay.²² XRN1 gene essentiality in cancer cell lines is correlated with the requirement for multiple genes beyond ADAR1, several of which encode proteins involved in RNA degradation and metabolism, including the RNA decapping proteins DCP2 and EDC4 (Figure 1C). Similar to ADAR1 gene essentiality,^7,8,12 analysis of gene expression in unmodified parental cancer cell lines predicted to require XRN1 showed an enrichment in gene sets that represent ISG expression (Figure 1D). To validate the results of this analysis, we used CRISPR-Cas9 to target control genes or XRN1 in cancer cell lines. CRISPR-Cas9 targeting resulted in efficient editing of control genes, with editing efficiencies ranging from 83% to 99%, and XRN1, with editing efficiencies ranging from 86% to 97% (Table S1), and robust depletion of XRN1 protein (Figure S1A). We found that lung cancer cell lines that require ADAR1 showed a significant cell viability defect after XRN1 deletion compared to control cells, whereas lung cancer cell lines that do not require ADAR1 also did not require XRN1 for their survival (Figure 1E). Next, we selected cancer cell lines from diverse lineages, including pancreatic adenocarcinoma, mesothelioma, triple-negative breast carcinoma, melanoma, and esophageal squamous cell carcinoma, which were predicted to require both ADAR1 and XRN1. XRN1 deletion in each of these cell lines caused a cell viability defect compared to controls (Figure 1F). Taken together, these data show that cancer cell lines that require ADAR1 also require XRN1 for survival, suggesting that these genes may function through similar pathways to control cancer cell viability.

Figure 1. The enzymatic activity of XRN1 is required for survival of cancer cell lines with high expression of ISGs.

(A) Correlation and statistical significance of genome-wide genetic dependencies compared to ADAR1 genetic dependency based on CRISPR-Cas9-mediated gene essentiality screens. Each dot represents a different gene. Pearson correlations and corresponding adjusted p (p_adj) values were computed for each feature in the Cancer Dependency Map Public 22Q4 dataset using all cancer cell lines.

(B) Correlation of ADAR1 and XRN1 genetic dependency based on CRISPR-Cas9-mediated gene essentiality screens. CHRONOS²³ is a computational method that predicts the sensitivity of cell lines to deletion of specific genes using data from gene essentiality screens. Lower CHRONOS scores predict for cell lines that are relatively sensitive to gene deletion, while higher scores predict for cell lines that are relatively insensitive to gene deletion. Each dot represents a different human cancer cell line. Pearson correlations were computed for each feature in the Cancer Dependency Map Public 22Q4 dataset using all cancer cell lines.

(C) Top genetic dependencies correlated with XRN1 genetic dependency in CRISPR-Cas9-mediated gene essentiality screens. Pearson correlations were computed for each feature in the Cancer Dependency Map Public 22Q4 dataset using all cancer cell lines.

(D) Expression gene set correlations with XRN1 genetic dependency based on CRISPR-Cas9-mediated gene essentiality screens. Pearson correlations and corresponding p_adj values were computed for each feature in the Cancer Dependency Map Public 22Q4 dataset using all cancer cell lines. A subset of the 200 genes with the highest absolute correlation was used to calculate gene set over-representation. C2 Canonical Pathways v7.0 and Hallmark v7.1 gene set collections from MSigDB were analyzed. p values were computed using a hypergeometric test and then corrected for multiple hypothesis testing using Benjamini-Hochberg false discovery rate correction to generate p_adj values. This analysis was generated from RNA-sequencing data from unmodified parental cancer cell lines.

(E) Lung cancer cell viability was assessed by ATP bioluminescence 12 days after targeting control loci (sg1: AAVS1; sg2: Chr2.2) or XRN1 (sg1: exon 6; sg2: exon 19) with CRISPR-Cas9. Each dot represents the average of three technical replicates from one of three independent biological replicates.

(F) Cell viability was assessed by ATP bioluminescence 12 days after targeting control loci or XRN1 with CRISPR-Cas9 in cancer cell lines of diverse lineages. Each dot represents the average of three technical replicates from one of three independent biological replicates.

(G) Cell viability was assessed by ATP bioluminescence (top) or crystal violet staining (bottom) 12 or 17 days, respectively, after targeting of a control locus or endogenous XRN1, by XRN1 sg1, with CRISPR-Cas9 in HCC366 cells expressing GFP control or overexpressing WT XRN1 resistant to XRN1 sg1 targeting or catalytically inactive mutants of XRN1 (D208A and E176G) resistant to XRN1 sg1 targeting. Each dot represents the average of three technical replicates from one of three independent biological replicates.

Raw data from gene essentiality screens in (A)–(D) were obtained from the Cancer Dependency Map.²¹ ATP bioluminescence values were normalized to the control sg1 sample within each cell line. Data from three independent biological replicates are shown in (E)–(G). Error bars represent standard deviation from the mean. *p < 0.05, **p < 0.01, ***p < 0.001, and ns = not significant, as calculated by repeated measures one-way analysis of variance (ANOVA) in (E) and (F) or paired Student’s t test in (G). See also Figure S1 and Table S1.

To determine whether the effects of XRN1 depletion are on target, we generated wild-type XRN1 constructs that are resistant to cleavage by XRN1 guides sg1 or sg2 and overexpressed these constructs in XRN1 KO-sensitive HCC366 and NCI-H1650 cells (Figures S1B–S1D). Overexpression of sg1/sg2-resistant wild-type XRN1 rescued deletion of endogenous XRN1 (Figures 1G and S1E–S1G). To test whether the enzymatic function of XRN1 is necessary for the survival of XRN1 KO-sensitive cancer cells, we generated sg1/sg2-resistant mutant versions of human XRN1 (D208A and E176G) that are orthologous to XRN1-inactivating mutations in K. lactis²⁴ and D. melanogaster.²⁵ Overexpression of XRN1 D208A and E176G failed to rescue cell viability in XRN1 KO-sensitive cancer cells after deletion of endogenous XRN1 (Figures 1G and S1E–S1G), demonstrating a requirement for XRN1 catalytic activity. Of note, HCC366 cells expressing the catalytically inactive versions of XRN1 (D208A and E176G) appear to have decreased cell viability compared to the GFP-expressing control cells after endogenous XRN1 deletion (Figures 1G and S1E), suggesting a possible dominant-negative function for these XRN1 mutations. However, we did not observe a similar effect in NCI-H1650 cells (Figures S1F and S1G), and HCC366 and NCI-H1650 cells expressing control single guide RNAs (sgRNAs) did not show a substantial growth defect (Figures 1G and S1E–S1G). Together, these data indicate that the enzymatic function of XRN1 is necessary for cancer cell survival in the setting of high ISG expression.

PKR is required for cancer cell lethality after XRN1 deletion

Since prior work showed that PKR contributes to ADAR1 genetic dependency in cancer cells^7,8,12 and that viral infection of XRN1-deficient cells stimulates PKR,^26,27 we examined whether PKR activation contributes to cancer cell lethality after XRN1 deletion. While XRN1 deletion did not increase PKR phosphorylation in XRN1 KO-insensitive cell lines, including A549 and NCI-H460 (Figure 2A), as well as NCI-H1299 and NCI-H1437 (Figure S2A), XRN1 loss led to increased PKR phosphorylation specifically in XRN1 KO-sensitive cell lines, including HCC366 and NCI-H1650 (Figure 2A), as well as HARA, HCC1438, and SW900 (Figure S2A).

Figure 2. PKR signaling is required for XRN1 gene essentiality.

(A) Representative immunoblots showing XRN1, phospho-PKR, total PKR, and β-actin protein levels in XRN1 KO-insensitive (A549 and NCI-H460) and XRN1 KO-sensitive (HCC366 and NCI-H1650) cells 7 days after CRISPR-Cas9 targeting of control loci or XRN1. At least three independent biological replicates were performed for each cell line. Molecular weight (MW) markers are shown in kilodaltons (kDa).

(B) Heatmaps showing relative transcripts per million (TPM) values for the indicated differentially expressed genes involved in the integrated stress response, the NF-κB pathway, and the interferon response pathway (rows) 7 days after CRISPR-Cas9 targeting of control loci or XRN1 (columns) in HCC366 cells. Red, increased expression; blue, decreased expression (as shown on the right side of the diagram). Three independent biological replicates were obtained for each condition.

(C) Immunofluorescence analysis of G3BP1 and Hoechst staining 7 days after CRISPR-Cas9 targeting of control loci or XRN1 in HCC366 cells (left panels). Scale bars, 20 μm. The percentage of cells containing at least one focus of G3BP1 staining was calculated in multiple fields across two independent biological experiments (right panel). Error bars represent standard deviation from the mean. *** p < 0.001, as calculated by one-way ANOVA.

(D–F) (Top panels) Immunoblots confirming depletion of PKR (D), MAVS (E), or RNase L (F) protein levels after CRISPR-Cas9 targeting in HCC366 cells. β-actin was used as a loading control. MW markers are shown in kDa. (Middle panels) Cell viability was assessed by ATP bioluminescence 12 days after CRISPR-Cas9 targeting of control loci or XRN1 in PKR-depleted (D), MAVS-depleted (E), or RNase L-depleted (F) HCC366 cells. ATP bioluminescence values were normalized to the control sg1 sample within each cell line. Each dot represents the average of three technical replicates from one of three independent biological replicates. Error bars represent standard deviation from the mean. ***p < 0.001 and ns = not significant, as calculated by repeated measures two-way ANOVA. (Bottom panels) Crystal violet staining 17 days after CRISPR-Cas9 targeting of control loci or XRN1 in PKR-depleted (D), MAVS-depleted (E), or RNase L-depleted (F) HCC366 cells. Crystal violet images are representative of three independent biological replicates.

See also Figure S2; Tables S2 and S3.

To determine whether gene expression pathways downstream of PKR are activated by XRN1 deletion, we performed RNA sequencing of the XRN1 KO-sensitive cell lines, HCC366 and NCI-H1650, after sgRNA targeting of control genes or XRN1. Principal component analysis (PCA) of RNA-sequencing data showed that control samples were generally most similar to one another, and XRN1 KO samples belonged to a distinct group in both cell lines, according to the major principal component, denoted as PC1 (Figures S2B and S2C). Of note, RNA sequencing of HCC366 cells was performed in two separate batches, which is consistent with the second principal component, PC2, being batch-associated (Figure S2B). Further analysis showed that XRN1 deletion in KO-sensitive cells resulted in increased expression of genes involved in the integrated stress response, the nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) pathway, and the interferon response pathway (Figures 2B and S2D; Table S2), all of which are known downstream targets of PKR signaling.¹⁷ Notably, the relative contributions of the genes displayed in Figures 2B and S2D are of significantly different magnitude for PC1 than for PC2, with Wilcoxon signed-rank test (paired) being 1.8 × 10⁻⁷ for HCC366 cells and 7.4 × 10⁻⁴ for NCI-H1650 cells (Table S2). These results are consistent with the conclusion that the major gene expression differences observed in these datasets are associated with XRN1 KO rather than other factors.

Moreover, XRN1 deletion in KO-sensitive cell lines caused increased stress granule formation as measured by G3BP1 relocalization, a marker of PKR activation,^28,29 compared to control cells (Figure 2C). Together, these data indicate that XRN1 deletion in XRN1 KO-sensitive cancer cells causes PKR pathway activation.

Next, we interrogated the requirement of PKR and other dsRNA sensing pathways for XRN1 genetic dependency. Deletion of XRN1 alone, PKR alone, or both XRN1 and PKR, followed by RNA-sequencing analysis, showed three distinct groups by PCA, corresponding to the three conditions (Figure S2E). Of note, PKR single KO samples were similar to control samples by PCA (Figure S2E). Co-deletion of PKR with XRN1 deletion largely reversed the gene expression changes induced by XRN1 deletion alone (Figure S2F and Table S3) and rescued cell viability in KO-sensitive cells (Figures 2D and S2G). In contrast, co-deletion of MAVS or RNase L with XRN1 deletion had no effect on cancer cell viability (Figures 2E, 2F, S2H, and S2I). These data demonstrate that XRN1 deletion causes activation of the dsRNA sensor PKR and that PKR is required for XRN1 KO sensitivity.

Inhibition of interferon signaling with the JAK inhibitor ruxolitinib can decrease PKR levels and rescue sensitivity to XRN1 loss

Since the interferon-inducible protein PKR is required for XRN1 KO-sensitivity, we reasoned that there may be a mechanistic link between elevated ISG expression and XRN1 genetic dependency. Indeed, the most significant gene expression correlates with XRN1 KO sensitivity across cancer cell lines are ISGs, as shown in Figure 1D, including PKR as the fifth most correlated gene to XRN1 KO sensitivity (Figure 3A). Immunoblotting also confirmed that PKR protein levels are slightly higher (average of 1.92-fold, p = 0.0025) in XRN1 KO-sensitive cell lines compared to KO-insensitive cell lines (Figure S3A). Thus, we hypothesized that activated interferon signaling may increase cellular PKR levels, thereby sensitizing cancer cell lines with high ISG expression to XRN1 deletion.

Figure 3. Inhibition of JAK1/2 signaling can modulate PKR levels and sensitivity to XRN1 deletion.

(A) Correlation of expression levels of individual genes with XRN1 genetic dependency based on CRISPR-Cas9-mediated gene essentiality screens. Each dot represents one gene, and the top ten gene expression correlates are labeled in red. Pearson correlations and corresponding p_adj values were computed for each feature in the Cancer Dependency Map Public 22Q4 dataset using all cancer cell lines.

(B) Representative immunoblots from two independent biological replicates showing phospho-STAT1, total STAT1, total PKR, MDA5, and β-actin protein levels in a panel of XRN1 KO-sensitive cancer cell lines treated with either DMSO control or ruxolitinib (1 μM) for 24 h. Molecular weight (MW) markers are shown in kDa.

(C, E, and G) Representative immunoblots showing XRN1, phospho-STAT1, total STAT1, phospho-PKR, total PKR, and β-actin protein levels in control and XRN1-deleted NCI-H1650 (C), HCC1438 (E), and SW900 (G) cells treated with either DMSO control or ruxolitinib (1 μM). MW markers are shown in kDa. Three independent biological replicates were performed for each cell line.

(D, F, and H) Cell viability was assessed by either ATP bioluminescence (left panels) or crystal violet staining (right panels) after CRISPR-Cas9 targeting of control loci or XRN1 in NCI-H1650 (D), HCC1438 (F), and SW900 (H) cells treated with DMSO control or ruxolitinib (1 μM). ATP bioluminescence values were normalized to the control sg1 sample within each cell line.

Each dot represents the average of three technical replicates from one of three independent biological replicates in (D), (F), and (H). Error bars represent standard deviation from the mean. *p < 0.05 and ***p < 0.001, as calculated by repeated measures two-way ANOVA. Crystal violet images are representative of three independent biological replicates. See also Figure S3.

To interfere with cellular interferon signaling, we utilized the selective JAK1/2 inhibitor ruxolitinib,³⁰ which caused a dose-dependent inhibition of STAT1 phosphorylation and downregulation of ISG protein levels in cancer cell lines with high basal activation of the interferon response pathway (Figures 3B and S3B). Of note, a subset of XRN1 KO-sensitive cancer cell lines displayed higher basal levels of STAT1 phosphorylation (Figure 3B).

To determine the contribution of JAK1/2-dependent interferon signaling to XRN1 KO sensitivity, we treated control and XRN1 KO cells with vehicle (DMSO) or ruxolitinib. Ruxolitinib treatment decreased total PKR levels and attenuated PKR phosphorylation after XRN1 deletion as compared to DMSO treatment specifically in a group of XRN1 KO-sensitive cell lines with higher levels of basal STAT1 phosphorylation (Figure 3B), namely NCI-H1650, HCC1438, and SW900 cells (Figures 3C, 3E, and 3G). In conjunction with decreases in total and phospho-PKR levels, ruxolitinib treatment largely rescued cell viability after XRN1 deletion in NCI-H1650, HCC1438, and SW900 cells (Figures 3D, 3F, and 3H).

Conversely, in XRN1 KO-sensitive cell lines with lower levels of basal STAT1 phosphorylation, specifically HCC366, KYSE-70, and MDA-MB-157 cells (Figure 3B), ruxolitinib treatment did not decrease total PKR levels and either attenuated PKR phosphorylation only modestly (HCC366), or not at all (HCC1438 and SW900), after XRN1 deletion compared to DMSO treatment (Figures S3C, S3E, and S3G). Consequently, ruxolitinib treatment did not rescue cell viability after XRN1 deletion in HCC366, KYSE-70, or MDA-MB-157 cells (Figures S3D, S3F, and S3H). These data suggest that for a subset of XRN1 KO-sensitive cancer cell lines with elevated baseline STAT1 phosphorylation, the lethality upon XRN1 deletion is rescued by disruption of interferon signaling via inhibition of JAK1/2.

Activation of interferon signaling can increase PKR levels and sensitivity to XRN1 loss

To extend the concept that interferon pathway activation can modulate XRN1 KO sensitivity, we tested whether stimulation of interferon signaling can induce vulnerability to XRN1 depletion in XRN1 KO-insensitive cells with low levels of basal ISG expression, similar to our previous report that interferon stimulation sensitizes cancer cells to ADAR1 depletion.⁷ Interferon-β stimulation of XRN1 KO-insensitive cell lines (A549, NCI-H460, NCI-H1299, and NCI-H1437) increased levels of STAT1 phosphorylation, total PKR, and phospho-PKR in A549 and NCI-H1299 cells (Figure 4A) as well as NCI-H460 and NCI-H1437 cells (Figure S4A). The combination of XRN1 KO with interferon-β treatment produced the greatest increase in PKR phosphorylation, which was associated with a mobility shift of phospho-PKR to a higher molecular weight (Figure 4A).

Figure 4. Interferon-β stimulation can increase PKR levels and sensitivity to XRN1 deletion.

(A) Representative immunoblots showing XRN1, phospho-STAT1, total STAT1, MDA5, phospho-PKR, total PKR, and β-actin protein levels in control or XRN1 KO A549 (left) or NCI-H1299 (right) cells after 24 h of treatment with vehicle control (sterile water) or interferon-β (10 ng/mL). Molecular weight (MW) markers are shown in kDa.

(B) Cell viability was assessed by ATP bioluminescence in control or XRN1 KO A549 (left) or NCI-H1299 (right) cells 5 days after treatment with vehicle control (sterile water) or the indicated concentrations of interferon-β. Each dot represents the average of three technical replicates from one independent experiment.

(C) Representative immunoblots showing XRN1, phospho-PKR, total PKR, and β-actin protein levels in control, XRN1 single KO, PKR single KO, or XRN1/PKR double KO (DKO) A549 (left) or NCI-H1299 (right) cells after 24 h of treatment with vehicle control (sterile water) or interferon-β (10 ng/mL). MW markers are shown in kDa.

(D) Cell viability was assessed by ATP bioluminescence in control, XRN1 single KO, PKR single KO, or XRN1/PKR double KO (DKO) A549 (left) or NCI-H1299 (right) cells 5 days after treatment with vehicle control (sterile water) or the indicated concentrations of interferon-β. Each dot represents the average of three technical replicates from one independent experiment.

Three independent biological replicates were performed for each cell line in (A)–(D). ATP bioluminescence values were normalized to the vehicle control sample for each isogenic cell line in (B) and (D). Error bars represent standard deviation from the mean. See also Figure S4.

Interferon-β stimulation of A549 and NCI-H1299 cells led to an increase in cell lethality of XRN1-deleted isogenic cells compared to cells expressing control sgRNAs (Figure 4B). This interferon-β-induced cell lethality requires PKR, as co-deletion of PKR and XRN1 (Figure 4C) rescued cell viability at multiple doses of interferon-β (Figure 4D). Although interferon-β treatment activated interferon signaling in the XRN1 KO-insensitive cell lines NCI-H460 and NCI-H1437, XRN1 deletion did not alter either PKR phosphorylation or cell viability substantially in these cell lines (Figures S4A and S4B). These latter data suggest that additional cell line-specific factors can modulate the response to interferon signaling and XRN1 deletion. Together, these data show that interferon signaling can induce sensitivity to XRN1 deletion in a PKR-dependent manner and provide further support for a mechanistic connection between high ISG expression and XRN1 genetic dependency in cancer cell lines.

Endogenous sense/anti-sense transcripts accumulate after XRN1 deletion

As we have shown above, the exoribonuclease activity of XRN1 is required for the survival of cancer cell lines with high ISG expression. In addition, interferon signaling, which regulates the abundance of the key dsRNA sensor PKR, is a critical modulator of XRN1 KO sensitivity. Synthesizing these observations, we hypothesized that XRN1 deletion may cause an increase in endogenous dsRNA ligands for PKR.

Thus, we examined whether XRN1 deletion affected classes of transcripts previously reported as PKR ligands, specifically mitochondrial RNA²⁰ and endogenous retroelement RNA.³¹ First, we did not detect any significant increases in mitochondrial RNA transcripts after XRN1 deletion (Table S2). Next, we found that XRN1 deletion led to increased transcript levels of multiple human endogenous retrovirus 9 (HERV9) subfamilies (Figures S5A and S5B; Table S4). However, these increased HERV9 transcripts were entirely dependent on PKR activity (Figures S5B and S5C; Table S4), suggesting that HERV9 transcripts likely do not initiate PKR activation in XRN1-deficient cells. Notably, most other endogenous retroelement subfamilies that were differentially expressed between control and XRN1 KO HCC366 cells showed a fold change of less than 2 (Table S4). These data provide evidence that neither mitochondrial RNA nor endogenous retroelements are the culprit dsRNA ligands that stimulate PKR activation in XRN1 KO-sensitive cancer cell lines.

An alternative mechanism for PKR activation is suggested by prior studies which showed that XRN1 regulates the formation and accumulation of complementary sense/anti-sense RNA transcripts in S. cerevisiae^32–34 and during viral infection.^26,27 To identify candidate dsRNA ligands for PKR, we performed strand-specific RNA sequencing of XRN1-deleted and control isogenic cell lines to determine whether XRN1 modulates endogenous sense/anti-sense transcript levels in XRN1 KO-sensitive HCC366 cells. Notably, PCA showed tight clustering of the samples from each experimental condition in HCC366 cells (Figure S2E).

We developed a computational pipeline to identify regions of the transcriptome that may contain complementary sense/anti-sense RNA transcripts and to quantify the abundance of these overlapping transcripts (Figure 5A). In brief, we defined loci containing overlapping sense/anti-sense transcripts as genomic regions that are covered by at least one RNA-sequencing read on both strands. We named the RNA transcripts mapping to these transcribed loci “complementary sense/anti-sense transcript pairs.” To enable standardized quantitative analysis of these complementary sense/anti-sense transcript pairs across different experimental conditions, we merged overlapping genomic regions from all experimental replicates across conditions into “union sets” of minimal genomic regions containing complementary sense/anti-sense transcripts. To quantify the abundance of complementary sense/anti-sense transcripts, we tabulated the total number of reads mapped within the merged minimal genomic regions of overlapping transcription. The smaller number of reads of the two strands was set as the number of complementary sense/anti-sense transcript pairs for the region. We excluded genomic regions with less than five complementary transcript pair reads in at least half of the samples to increase confidence that these genomic regions did indeed contain overlapping transcripts.

Figure 5. Endogenous sense/anti-sense transcripts accumulate after XRN1 deletion.

(A) Schematic of computational pipeline to identify genomic regions containing complementary sense/anti-sense transcript pairs from strand-specific RNA sequencing data. We defined loci containing overlapping sense/anti-sense transcripts as genomic regions that are covered by at least one RNA sequencing read on both strands. We named the RNA transcripts derived from these loci as “complementary sense/anti-sense transcript pairs.” We merged overlapping genomic regions from all experimental replicates across different experimental conditions into “union sets” of minimal genomic regions containing complementary sense/anti-sense RNA transcripts (shown as “complementary region” in the schematic). To quantify the abundance of these complementary sense/anti-sense RNA transcripts, we tabulated the total number of reads in these merged minimal genomic regions. The smaller number of the reads of the two strands was set as the number of complementary sense/anti-sense transcript pairs for the region. We excluded genomic regions with less than five complementary pair reads in at least half of the samples. See also STAR Methods for additional details. Blue lines represent transcripts from the (−) DNA strand, and red lines represent transcripts from the (+) DNA strand. The (+) DNA strand is defined as the DNA strand with its 5′ end at the tip of the short arm (p arm) of each chromosome. The (−) DNA strand is defined as the DNA strand with its 5′ end at the tip of the long arm (q arm) of each chromosome.

(B) Volcano plot of differentially expressed complementary sense/anti-sense transcript pairs in XRN1 KO compared to control HCC366 cells. RNA extraction was performed 7 days after CRISPR-Cas9 targeting of control loci or XRN1. The horizontal red line corresponds to a p_adj value of 0.1. Each dot represents the complementary sense/anti-sense transcript pairs contained in one genomic region. These data represent an analysis of three independent biological replicates of control HCC366 cells and six independent biological replicates of XRN1 KO HCC366 cells.

(C) Quantitation of relative levels of complementary sense/anti-sense transcript pairs in control, XRN1 single KO, PKR single KO, and XRN1/PKR double KO (DKO) HCC366 cells. Green arrows show the relative quantity of “PKR-regulated” complementary pairs, and purple arrows show the relative quantity of “PKR-independent” complementary pairs. RNA extraction was performed 7 days after CRISPR-Cas9 targeting of a control locus, XRN1, PKR, or both XRN1 and PKR. These data represent an analysis of three independent biological replicates for control and PKR single KO HCC366 cells and six independent biological replicates for XRN1 single KO and XRN1/PKR double KO (DKO) HCC366 cells.

(D) Schematic depicting the relationship of the dsRNA sensor PKR to “PKR-independent” and “PKR-regulated” complementary sense/anti-sense transcript pairs identified in (C).

(E and F) Representative pileups from strand-specific RNA sequencing of control, XRN1 single KO, PKR single KO, and XRN1/PKR double KO (DKO) HCC366 cells showing examples of “PKR-independent” and “PKR-regulated” complementary sense/anti-sense transcript pairs. Strand-specific RNA sequencing data obtained from the indicated isogenic cell lines are displayed, with blue pileup traces representing transcripts from the (−) DNA strand and red pileup traces representing transcripts from the (+) DNA strand. The (+) DNA strand is defined as the DNA strand with its 5′ end at the tip of the short arm (p arm) of each chromosome. The (−) DNA strand is defined as the DNA strand with its 5′ end at the tip of the long arm (q arm) of each chromosome. Annotated transcripts associated with each locus are shown at the bottom and are color-coded according to the DNA strand from which they are transcribed. The arrows at the bottom show the annotated direction of transcription.

(G) Quantitation of relative levels of “PKR-independent” complementary sense/anti-sense transcript pairs in control, XRN1 single KO, PKR single KO, and XRN1/PKR double KO (DKO) HCC366 cells. These data represent an analysis of three independent biological replicates for control and PKR single KO HCC366 cells and six independent biological replicates for XRN1 single KO and XRN1/PKR double KO (DKO) HCC366 cells.

Relative values of complementary sense/anti-sense transcript pairs were normalized to the control HCC366 sample. Error bars represent standard deviation from the mean. ***p < 0.001 and ns = not significant, as calculated by two-sided Student’s t test. See also Figure S5; Tables S4 and S5.

Transcriptome-wide analysis showed that targeting the XRN1 KO-sensitive HCC366 cell line with XRN1 sgRNA caused a significant increase in complementary sense/anti-sense transcript pairs in over 900 genomic regions, as compared to control sgRNA (Figure 5B and Table S5). These regions containing complementary sense/anti-sense transcript pairs can span many kilobases of DNA (Figure S5D), although we do not know from short-read sequencing assemblies whether these represent contiguous transcripts in all cases. These genomic regions contain combinations of protein-coding genes (both exons and introns), non-coding RNAs, and pseudogenes (Figure S5E). We cannot determine from these RNA-sequencing data whether XRN1 degrades sense/anti-sense transcripts in their single-stranded form or in a double-stranded form. The 5′ to 3′ processivity of XRN1 would suggest that double-stranded sense/anti-sense transcript pairs would need the 5′ ends of each transcript to be exposed to allow for XRN1-mediated degradation.²⁵ To test this notion, we compiled all complementary sense/anti-sense transcript pairs that accumulate in HCC366 cells after XRN1 depletion in which transcripts derived from both strands are annotated (~12% of all complementary sense/anti-sense transcript pairs). We found that the corresponding annotations for these overlapping transcript pairs are oriented preferentially in a configuration with their 5′ ends exposed for degradation by XRN1 rather than the configuration where the 5′ ends are protected from XRN1 degradation (Figure S5F and Table S5), as determined by the comprehensive cis-natural anti-sense transcripts identifier via RNA-sequencing data (CCIVR2) computational tool.³⁵ The exposed 5′ ends of these annotated overlapping transcripts also contain at least four “overhanging” base pairs in 98.0% of 150 transcripts (Figure S5G and Table S5), consistent with findings from RNA:DNA hybrid substrates that are susceptible to XRN1 degradation.²⁵ These data suggest that XRN1 deletion in HCC366 cells causes accumulation of complementary sense/anti-sense transcripts that are normally degraded by XRN1.

Next, we investigated whether PKR is required for accumulation of these complementary transcript pairs. We found that co-deletion of both PKR and XRN1 caused a smaller increase in complementary sense/anti-sense transcripts compared to XRN1 deletion alone (Figure 5C and Table S5). This result suggests that there could be two classes of XRN1-regulated complementary RNA pairs: one class that is independent of PKR signaling (“PKR-independent,” shown in purple) and a second class that is regulated directly by PKR activation (“PKR-regulated,” shown in green) (Figure 5D and Table S5). Among complementary sense/anti-sense transcripts classified as PKR independent, the SLC7A6/SLC7A6OS (Figure 5E) and SZT2/HYI (Figure S5H) loci represent two of the top ten most abundant RNA pairs that are significantly increased after XRN1 deletion. Likewise, among complementary sense/anti-sense transcripts classified as PKR regulated, the C11orf68 (Figure 5F) and CFLAR/CFLAR-AS1 (Figure S5I) loci represent two of the top ten most abundant RNA pairs that are significantly increased after XRN1 deletion. We reasoned that the PKR-independent complementary RNA transcripts may function as the RNA ligands that initially activate PKR in XRN1-deficient cancer cells. Notably, we found significantly increased levels of PKR-independent sense/anti-sense transcript pairs in XRN1-depleted HCC366 cells as compared to HCC366 cells expressing control sgRNA (Figure 5G). These data suggest that XRN1 regulates the accumulation of endogenous complementary sense/anti-sense transcript pairs in human cancer cells, likely through both direct RNA degradation and modulation of PKR activation. Furthermore, these endogenous transcript pairs have the potential to form dsRNA through base-pair complementarity and may represent candidate ligands for PKR in cancer cells.

DISCUSSION

Growing evidence suggests that targeting innate immune pathways in cancer cells, including regulators of dsRNA metabolism, may represent a promising therapeutic strategy.¹⁴ Previous work from our group and others has identified the RNA editing enzyme, ADAR1, as a unique genetic dependency in a subset of cancer cells with interferon response pathway activation.^7,8,12 These studies have also implicated a critical role for the dsRNA sensor PKR in modulating cancer cell survival after ADAR1 inactivation.

Here, we show that the 5′ to 3′ exoribonuclease XRN1 is a selective genetic dependency in human patient-derived cancer cells, with XRN1 dependency correlated with activation of the interferon response pathway. We demonstrate further that human cancer cell lines with an activated interferon cell state tend to have higher levels of the dsRNA sensor PKR and that PKR is required for XRN1 genetic dependency.

We propose that XRN1 activity allows cancer cells to tolerate activation of the interferon response pathway as described in the following model (see graphical abstract). One role for XRN1 is to degrade anti-sense RNA transcripts, thereby preventing formation of complementary sense/anti-sense RNA pairs and limiting activation of the PKR dsRNA sensing pathway. This proposed mechanism is concordant with observations in yeast, where XRN1-sensitive unstable transcripts can form dsRNA structure.^33,34 Depletion of XRN1 in interferon-activated cancer cells causes accumulation of complementary sense/anti-sense RNA pairs, which may function as dsRNA ligands for PKR that trigger PKR pathway activation and cancer cell lethality. Disruption of the activated interferon cell state with the JAK1/2 inhibitor ruxolitinib can decrease cellular PKR levels and rescue cell lethality induced by XRN1 depletion. Conversely, stimulation of cancer cell lines with lower expression of ISGs with exogenous interferon-β can increase cellular PKR levels and induce sensitivity to XRN1 depletion. Overall, our data suggest that the abundance of both dsRNA ligand and PKR levels may contribute to XRN1 KO sensitivity in cancer cells with interferon pathway activation.

We showed that XRN1 regulates the accumulation of endogenous complementary sense/anti-sense transcript pairs, which may represent candidate PKR ligands. Studies using crosslinking and immunoprecipitation followed by RNA sequencing (CLIP-seq) analysis of PKR in XRN1-sufficient cells have identified bidirectionally transcribed mitochondrial RNAs and inverted Alu retroelements as bona fide PKR ligands.²⁰ Whether PKR is activated by specific complementary RNA pairs or a general accumulation of sequence-agnostic complementary RNA pairs in XRN1-depleted cancer cells remains an open question. Answering this complex question is likely to require the simultaneous targeting of multiple genomic loci that encode complementary RNA pairs, requiring the development of novel experimental methods.

Viewing the regulation of dsRNA more broadly, a recent study showed that sense/anti-sense RNAs may be potential substrates for human ADAR1,³⁶ which is also a negative regulator of PKR. In addition, analysis of genome-wide association studies showed that a subset of these ADAR1-edited complementary RNA pairs was transcribed at genomic loci implicated in the pathogenesis of a range of autoimmune diseases.³⁶ Thus, future studies may reveal additional roles for complementary sense/anti-sense RNA pairs in cancer and autoimmune disease.

A fraction of the endogenous complementary sense/anti-sense transcript pairs we identified in XRN1-deficient cells were PKR regulated. Prior studies have shown that DNA virus infection can trigger host RNA polymerases to transcribe immunogenic dsRNA from the viral genome,^37,38 thereby amplifying the antiviral innate immune response. Similarly, our data suggest that XRN1 deletion can activate PKR to cause accumulation of additional PKR-regulated complementary sense/anti-sense RNA pairs. These PKR-regulated complementary transcript pairs may serve as a positive feedback mechanism to amplify PKR activation and may also have the potential to stimulate additional dsRNA sensors. Whether PKR signaling can generate “endogenous adjuvant” signals for the innate immune system merits additional investigation.

Our work suggests that interferon pathway activation can induce XRN1 KO sensitivity, at least in part, by controlling the abundance of the interferon-inducible protein PKR. Of note, several studies have demonstrated that a subset of human cancer cell lines and patient tumors of diverse lineages exhibit cancer cell-autonomous interferon pathway activation.^8,39,40 From a therapeutic perspective, our study suggests that interferon pathway activation in cancer cells may create cell state-specific genetic vulnerabilities whereby targeting molecular regulators of dsRNA sensors, such as PKR, can stimulate a viral mimicry response to promote cancer cell lethality. Aside from genetic vulnerabilities, one study showed that the activated interferon cell state enhances the response to anthracycline-based chemotherapy in cancer cell lines and patients with breast cancer.³⁹ Identification of additional genetic or pharmacologic vulnerabilities associated with the activated interferon cell state may improve therapeutic targeting of this subset of human cancers.

In addition to showing that XRN1 inactivation can induce direct cancer cell killing, our data may have implications for overcoming resistance to cancer immunotherapy. Multiple studies have demonstrated that resistance to immune checkpoint inhibition is associated with activation of an ISG expression program in cancer cells.^41–43 While an elevated ISG expression program may confer an immunotherapy-resistant cell state, activated interferon signaling may also increase the abundance of PKR and create a cell state-specific vulnerability to XRN1 inhibition. A recent study showed that XRN1 depletion in mouse tumor models can potentiate the effect of cancer immunotherapy through a mitochondrial antiviral-signaling protein (MAVS)-dependent mechanism.⁴⁴ While the mouse cancer cells tested in that study do not require XRN1 for survival, future work may clarify the precise role of PKR in the response of XRN1-depleted cells to cancer immunotherapy.

Taken together, our study provides mechanistic insight into how XRN1 regulates the activity of the dsRNA sensor PKR in cancer cells. Moreover, our data show how a cell-intrinsic activated interferon cell state can create unique genetic vulnerabilities in human cancer cells. Future studies will be required to determine whether XRN1 can be targeted specifically in cancer cells.

Limitations of the study

This study identifies XRN1 as a genetic vulnerability in human cancer cell lines with high expression of ISGs. Our data suggest that XRN1 deletion causes accumulation of complementary sense/anti-sense RNA transcript pairs that may function as ligands for the dsRNA sensor PKR. Additional experiments using methods such as CLIP-seq are required to determine whether these complementary sense/anti-sense transcripts can bind to PKR, which would strengthen the notion that these RNAs are PKR ligands. Thus, future studies that seek to determine whether complementary sense/anti-sense RNA transcript pairs are bona fide PKR ligands may elucidate the mechanism by which XRN1 regulates the dsRNA sensor PKR.

In addition, although it is tempting to consider whether inhibition of XRN1 may be an effective therapeutic strategy in human cancer, Xrn1 deletion in mice results in embryonic lethality,³⁵ suggesting that XRN1 inhibition may cause substantial toxicity. Thus, XRN1 would need to be targeted specifically or at least preferentially in cancer cells. Whether such a therapeutic window exists for inhibition of XRN1 in cancer cells may help illuminate the potential clinical utility of the findings in this study.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for reagents may be directed to and will be fulfilled by the lead contact, Matthew Meyerson (matthew.meyerson@dfci.harvard.edu).

Materials availability

All plasmids generated in this study will be deposited in Addgene upon publication. Any additional materials will be available upon publication and request.

Data and code availability

• The raw and processed bulk RNA sequencing data presented in this manuscript have been deposited at the NCBI Gene Expression Omnibus (GEO) repository and are publicly available as of the date of publication. Accession numbers are listed in the key resources table. The source data used to generate Figures 1A–1D and 3A were obtained from publicly available datasets from the Cancer Dependency Map Portal [http://depmap.org/portal/]. Immunoblotting and microscopy data reported in this paper will be shared by the lead contact upon request.

• All original code in this manuscript has been deposited at GitHub and is publicly available. The DOI to access the original code is listed in the key resources table.

• Any additional information required to re-analyze the data reported in this paper will be available from the lead contact upon request.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit monoclonal anti-phospho-PKR Thr446	Abcam	Cat# ab32036; RRID: AB_777310
Rabbit anti-RNase L antibody	Abcam	Cat# ab191392; RRID: AB_2916269
Rabbit polyclonal anti-XRN1	Bethyl Laboratories	Cat# A300-443A; RRID: AB_2219047
Rabbit monoclonal anti-ISG15	Cell Signaling Technology	Cat# 2758; RRID: AB_2126200
Rabbit monoclonal anti-G3BP1	Cell Signaling Technology	Cat# 17798; RRID: AB_2884888
Rabbit polyclonal anti-MAVS	Cell Signaling Technology	Cat# 3993; RRID: AB_823565
Rabbit monoclonal anti-MDA5	Cell Signaling Technology	Cat# 5321; RRID: AB_10694490
Rabbit polyclonal anti-total PKR	Cell Signaling Technology	Cat# 3072; RRID: AB_2277600
Rabbit monoclonal anti-phospho-STAT1 (Tyr701)	Cell Signaling Technology	Cat# 9167; RRID: AB_561284
Rabbit polyclonal anti-total STAT1	Cell Signaling Technology	Cat# 9172; RRID: AB_2198300
Rabbit polyclonal anti-XRN1	Cell Signaling Technology	Cat# 70205; RRID: AB_2799779
Mouse monoclonal anti-β-actin	Santa Cruz Biotechnology	Cat# sc-47778; RRID: AB_626632
Goat anti-mouse IRDye 680LT	LI-COR Biosciences	Cat# 926-68020; RRID: AB_10706161
Goat anti-rabbit IRDye 800CW	LI-COR Biosciences	Cat# 926-32211; RRID: AB_621843
Alexa Fluor® 647 AffiniPure Donkey anti-Rabbit IgG	Jackson ImmunoResearch Laboratories	Cat# 711-605-152; RRID: AB_2492288
Chemicals, peptides, and recombinant proteins
Blasticidin	Thermo Fisher Scientific	Cat# A1113903
cOmplete™ Mini Protease Inhibitor Cocktail Tablets	Roche Diagnostics	Cat# 4693124001
Human IFN-Beta 1a, mammalian	PBL Assay Science	Cat# 114151
Laemmli SDS Sample Buffer, reducing (6X)	Thermo Fisher Scientific	Cat# AAJ61337-AC
Polybrene	Santa Cruz Biotechnology	Cat# sc-134220
PhosSTOP™	Roche Diagnostics	Cat# 4906837001
Puromycin	Thermo Fisher Scientific	Cat# A1113803
Restore PLUS Western Blot Stripping Buffer	Thermo Fisher Scientific	Cat# PI46430
RIPA Buffer	Sigma Aldrich	Cat#R0278
Ruxolitinib	Selleck Chemicals	Cat#S1378
Critical commercial assays
Pierce BCA Protein Assay Kit	Thermo Fisher Scientific	Cat# 23225
CalPhos™ Mammalian Transfection Kit	Takara Bio	Cat# 631312
CellTiter-Glo® Luminescent Cell Viability Assay	Promega	Cat# PAG7572
NEBNext rRNA Depletion Kit	New England Biolabs	Cat#E6310
NEBNext Ultra II Directional RNA Library Prep Kit	New England Biolabs	Cat#E7760
QIAseq FastSelect -rRNA HMR Kit	Qiagen	Cat# 334375
RNeasy Plus Kit	Qiagen	Cat# 74134
Deposited data
Raw data from bulk RNA-sequencing	This manuscript	NCBI Gene Expression Omnibus (GEO) Accession Number GEO: GSE248036
Experimental models: Cell lines
A549 (Control, XRN1 single KO, PKR single KO, and XRN1/PKR double KO)	This manuscript	N/A
NCI-H1299 (Control, XRN1 single KO, PKR single KO, and XRN1/PKR double KO)	This manuscript	N/A
Oligonucleotides
sgRNA sequences for CRISPR-Cas9-mediated gene knockout	See Table S5	N/A
Recombinant DNA
plentiCRISPRv2 with puromycin resistance	Addgene	Cat# 98290; RRID: Addgene_98290
plentiCRISPRv2 with blasticidin resistance	Cloned in-house, adapted from Addgene plasmid	N/A
plentiCRISPRv2 expressing two sgRNAs with puromycin resistance	Cloned in-house, adapted from Addgene plasmid	N/A
pLEX307	Addgene	Cat# 41392; RRID: Addgene_41392
pMD2.G	Broad Institute of Harvard and MIT	N/A
psPAX2	Broad Institute of Harvard and MIT	N/A
pLNHA-C1-HsXRN1 (XRN1 ORF)	Addgene	Cat# 66596; RRID: Addgene_66596
XRN1 D208A ORF, resistant to XRN1 sgRNA targeting	Cloned in-house, adapted from Addgene plasmid	N/A
XRN1 E176G ORF, resistant to XRN1 sgRNA targeting	Cloned in-house, adapted from Addgene plasmid	N/A
Software and algorithms
Custom code for identification of complementary sense/anti-sense RNA pairs	This manuscript	https://doi.org/10.5281/zenodo.7971843

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell lines

A549 (Male), NCI-H460 (Male), NCI-H1299 (Male), NCI-H1437 (Male), HARA (Male), HCC366 (Female), HCC1438 (Male), NCI-H1650 (Male), KYSE-70 (Male), MDA-MB-157 (Female), and NCI-H2052 (Male) cells were cultured in RPMI medium supplemented with 10% fetal bovine serum. A375 (Female), PATU-8902 (Female), and human embryonic kidney (HEK) 293T (Female) cells were cultured in DMEM supplemented with 10% fetal bovine serum. All cells were cultured at 37°C with 5.0% carbon dioxide.

All cancer cell lines were obtained from the Broad Institute Cancer Cell Line Encyclopedia⁴⁵ (CCLE) or the American Type Culture Collection (ATCC). Both CCLE and ATCC perform cell line authentication and mycoplasma testing routinely. All cell lines were tested for mycoplasma prior to use and regularly thereafter.

METHOD DETAILS

CRISPR-Cas9 gene knockout

Single guide RNA sequences were designed using the sgRNA Designer tool (The Broad Institute Genomics Perturbation Platform) (https://portals.broadinstitute.org/gpp/public/analysis-tools/sgrna-design). sgRNA sequences are displayed in Table S1. sgRNAs were cloned into the Cas9 expressing lentiviral vector lentiCRISPRv2 or a modified lentiCRISPRv2 construct that expresses two different sgRNAs under the control of separate human and mouse U6 promoters. Individual lentiCRISPRv2 vectors were introduced along with packaging vectors into HEK 293T cells via calcium phosphate transfection according to the manufacturer’s protocol (Clontech). Lentivirus was harvested at 48 and 72 h after transfection in RPMI medium supplemented with 10% FBS and filtered with 45 μm filters prior to transduction of cancer cell lines using centrifugation at 1000g for 2 h in the presence of 8 μg/mL of polybrene (Santa Cruz Biotechnology). Transduced cell lines were selected in puromycin and/or blasticidin (Thermo Fisher Scientific) for at least 5 days prior to use in assays. Protein lysates were collected from the transduced cells and protein levels of the targeted gene(s) were assessed by immunoblotting to validate gene KO.

Cell viability assays

Cell counting was performed using a Coulter Particle Counter (Beckman-Coulter). For ATP bioluminescence experiments, cells were plated at a density of 1,500 (HARA, HCC1438, NCI-H2052, PATU-8902) or 3,000 (A375, A549, KYSE-70, MDA-MB-157, NCI-H460, NCI-H1299, NCI-H1437, HCC366, NCI-H1650) cells per well in 96 well assay plates (Corning). ATP bioluminescence was assessed at 12 days after gene KO with the CellTiter-Glo Luminescent Cell Viability Assay (Promega). For crystal violet staining assays, cells were plated at a density of 10,000 or 20,000 cells per well in 12 well tissue culture plates. Once the control cells grew to near confluency, each well was washed twice with ice-cold PBS, fixed on ice with ice-cold methanol for 10 min, stained with 0.5% crystal violet solution (made in 25% methanol) for 10 min at room temperature, and washed at least four times with water. All cell viability assays were performed in at least technical and biological triplicates.

XRN1 mutagenesis and overexpression

An XRN1 open reading frame (ORF) clone deposited by Elisa Izaurralde was purchased from Addgene (#66596). The entire ORF was sequenced to confirm fidelity to the NCBI Reference Sequence NM_019001.5. An entry clone for XRN1 was obtained through PCR-amplification of the XRN1 ORF and was sub-cloned into a Gateway donor vector. Overlap PCR was performed to introduce silent mutations separately into the protospacer adjacent motif (PAM) sequences targeted by XRN1 sgRNA 1 and sgRNA 2, thereby rendering the constructs resistant to CRISPR-Cas9 editing by these sgRNAs. Subsequently, the D208A and E176G mutations were engineered into each XRN1 sg1 and sg2-resistant XRN1 construct separately using overlap PCR. The resulting XRN1 sg1 and sg2-resistant XRN1 constructs and a GFP control construct were sub-cloned into the pLEX307 lentiviral expression vector (Addgene) under the control of an EF-1α promoter. Each expression vector was then transfected into HEK 293T cells to generate lentivirus. Lentiviral transduction of target cell lines was performed as described above.

Immunoblotting

Cells were lysed in RIPA lysis buffer (Thermo Fisher Scientific) supplemented with 1× protease and phosphatase inhibitor cocktails (Roche). Protein concentrations were obtained using the BCA Protein Assay Kit (Pierce) and 6X Laemmli SDS sample buffer (Thermo Fisher Scientific) was added to protein extracts. Protein extracts were normalized between all samples within an experiment and boiled above 95°C for 10 min. Proteins were resolved on 4–12% Bis-Tris gradient gels, transferred to nitrocellulose membranes, and immunoblotting with primary and secondary antibodies was performed according to standard procedures. All primary antibodies were used at a dilution of 1:1000 except anti-β-actin, which was used at a dilution of 1:10,000. Secondary antibodies from LI-COR Biosciences were used at a dilution of 1:10,000. Selected immunoblots were stripped with Restore PLUS Western Blot Stripping buffer (Thermo Fisher Scientific) prior to repeat immunoblotting. Immunoblots were imaged using the LI-COR digital imaging system. Quantitation of band intensities was performed with ImageJ. All immunoblots were cropped to optimize clarity and presentation.

Fluorescence microscopy

Control and XRN1 KO HCC366 cells were seeded on coverslips to 60–80% confluency. Cells were fixed with 4% paraformaldehyde for 10 min at room temperature, washed with PBS, and permeabilized with 0.2% Triton X-for 10 min at room temperature. Fixed and permeabilized cells were incubated with phosphate buffered saline with Tween (PBS-T) containing 1% bovine serum albumin for 30 min at room temperature and washed with PBS-T. Samples were stained using primary antibody for 1 h at room temperature, washed twice with PBS-T, stained with secondary antibody for 1 h at room temperature, and then washed twice with PBS-T. Nuclei were stained with Hoechst 33342 for 5 min at room temperature. Coverslips were mounted using Fluoromount-G and imaged with an Olympus Fluoview FV10i confocal microscope at 60X magnification with an oil-immersion lens. Microscopy images were processed with ImageJ.

Ruxolitinib treatment

Unmodified parental cancer cell lines were plated in 6 well plates and treated with DMSO or ruxolitinib the next day. After 24 h of ruxolitinib treatment, protein lysates for each parental cell line were harvested as described above. Control and XRN1 KO cancer cell lines were treated with DMSO or ruxolitinib starting at 24 h after lentiviral transduction. Culture medium containing DMSO or ruxolitinib was changed at least once every 3 days until protein lysates were harvested or cell viability measurements were obtained.

Interferon treatment

For interferon treatment assays, cells were plated at a density of 3,000 cells per well in a 96 well assay plate (Corning). The following day, cells were treated with increasing doses of recombinant human IFN-beta 1a (mammalian) protein (PBL Assay Science) or sterile water as the vehicle control. Cell viability was assessed 5 days after interferon-β treatment with the CellTiter-Glo Luminescent Cell Viability Assay (Promega). ATP bioluminescence values of the interferon-β treated wells were normalized to those of the corresponding vehicle-treated controls. Dose curves were obtained using nonlinear regression on a standard four-parameter logistic model using GraphPad Prism 7.

Sample preparation and RNA sequencing

RNA was isolated from HCC366 and NCI-H1650 cells 7 days after transduction with lentivirus co-expressing Cas9 and sgRNAs targeting control loci, XRN1, PKR, or both XRN1 and PKR. Total RNA was isolated using the RNeasy Plus Kit (Qiagen) followed by ribosomal RNA depletion using the NEBNext rRNA Depletion Kit (New England Biolabs) or the QIAseq FastSelect -rRNA HMR Kit (Qiagen). RNA sequencing libraries were prepared using the NEBNext Ultra II Directional RNA Library Prep Kit (New England Biolabs) and sequenced on an Illumina HiSeq instrument (150-bp paired-end reads).

Processing of RNA sequencing data

Quality check of RNA sequencing FASTQ files was completed using FastQC v0.11.9 with default parameters. After manual inspection, it was determined that the first three nucleotides of all reads should be removed before mapping. Adapter trimming was completed using cutadapt v2.8; Illumina Universal Adapter sequence “AGATCGGAAGAG” was used for both ends. After trimming, reads shorter than 35 bp were discarded. Reads were mapped to the Ensembl v99 hg38 gene annotation using the STAR aligner v2.7.5b. Read pileup was extracted using STAR.

Differential gene expression analysis of RNA sequencing data

RSEM⁴⁶ was run on aligned BAM files with parameters –estimate-rpsd –seed 12345 –strandedness reverse to compute TPMs for each gene. PCA plots of the gene expression data were generated with DESeq2⁴⁷ using TPM values to assess associations between samples within and between each experimental condition. Gene expression heatmaps were generated using TPM values and the R package pheatmap [https://cran.r-project.org/web/packages/pheatmap/index.html]. DESeq2⁴⁷ was used to evaluate differential gene expression by estimation of fold change and dispersion of data. The cutoff for differential expression between two groups was a false discovery rate (FDR) < 0.1. Fold changes for each gene were calculated by the average log2 expression value across groups, where positive values represent upregulation and negative values represent downregulation of gene expression. To determine whether the selected sets of differentially expressed genes contribute differently to PC1 and PC2 of the PCA, a paired Wilcoxon signed-rank test was performed.

Identification of endogenous retroelements in RNA sequencing data

We used the SQuIRE⁴⁸ computational pipeline to quantify endogenous retroelements at the subfamily level. SQuIRE was run with default parameters with the steps including Build, Fetch, and Map. The Count step was run twice for each sample, with the strandedness parameters set to 1 and 2 to obtain the read counts on the (+) and (−) DNA strands, respectively.

Transcriptome-wide identification of complementary sense/anti-sense transcript pairs

We defined bidirectionally transcribed loci as genomic regions that are covered by at least one RNA sequencing read on both strands. In this step, pair-end reads were converted to fragments by joining the two mates and setting its orientation following the first mate. If one of the two mates was not mapped, the other mate was still included in the analysis. For simplicity, all such mapped read singletons or fragments are referred to as “reads” in this section. To be considered as a bidirectionally transcribed locus, a region must be supported by at least two reads, one mapped to the (+) DNA strand and the other to the (−) DNA strand. We named the RNA transcripts derived from these bidirectionally transcribed loci as “complementary sense/anti-sense transcript pairs.”

Many of these regions resulted from overlapping portions of only two reads, leading to a heavily fragmented bidirectional transcriptome for any one RNA sequencing sample. To enable effective quantitative analysis between different experimental conditions, we constructed union sets of these regions across multiple samples. Regions belonging to samples from the same cancer cell line were used to construct one union set by merging all overlapping regions. We then quantified the abundance of bidirectional transcription in each sample based on the number of reads mapped within the range of each merged region.

Quantifying the abundance of complementary sense/anti-sense transcript pairs

To represent the abundance of bidirectional transcription within a genomic region, we defined a metric termed complementary pairs (CP). A complementary pair consists of two overlapping reads that were mapped to the (+) and (−) DNA strands, respectively. The CP value of a genomic region was set as equal to the smaller of the number of reads mapped to the (+) DNA strand or the number of reads mapped to the (−) DNA strand.

After constructing the union set of each cell line, we summed the CP values of all sub-regions from the same sample to obtain the CP value for the merged genomic region. The resulting CP values were used as read count inputs for DESeq2 analysis to detect differential bidirectional transcription across experimental conditions within the same cell line. A count matrix was constructed with columns representing the samples and rows representing the merged genomic regions. To reduce the number of fragmented regions, we excluded genomic regions with fewer than 5 CP in at least half of the samples. DESeq2 was run with default parameters, and genomic regions with a p_adj value less than 0.1 and a log2 fold change with an absolute value greater than 1 were considered as significant.

Determination of the length and transcript composition of complementary sense/anti-sense transcript pairs

To characterize the genomic regions of bidirectional transcription, we performed additional analysis of the complementary sense/anti-sense transcript pairs that showed a significant increase (log2-fold change greater than 0, p_adj less than 0.1) after XRN1 KO and XRN1/PKR double KO in HCC366 cells. Lengths of genomic regions were calculated by obtaining the genomic distance, in base pairs, from the two ends of each genomic region. Transcripts mapping within these genomic regions and their corresponding types were retrieved using Ensembl human v.105 (annotation hub ‘AH98047’). Genomic regions were categorized as containing either protein-coding transcripts only, non-coding transcripts (encompassing lncRNAs, miRNAs, intergenic transcripts, etc.) only, pseudogene transcripts, or combinations of these three categories of transcripts.

Determination of the configuration and 5′ overhang length of complementary sense/anti-sense transcript pairs

Annotated complementary sense/anti-sense transcript pairs that were increased after XRN1 KO and/or XRN1/PKR double KO were analyzed using the CCIVR2³⁵ tool to determine the configuration of these overlapping transcripts, either with their 5′ or 3′ ends exposed. To determine the length of the 5′ overhang of annotated overlapping transcript pairs with exposed 5′ ends, as determined by the CCIVR2 tool, we generated individual BAM files for transcripts derived from both (+) and (−) DNA strands using SAMtools v1.6. The 5′ overhang region for each transcript derived from the (+) DNA strand was demarcated by the annotated 5′ end position of the transcript and the annotated 3′ end position of the overlapping transcript pair derived from the (−) DNA strand. The 5′ overhang length of each complementary transcript pair was calculated by subtracting the most 3′ position of the transcript reads derived from the (−) DNA strand from the most 5′ position of the transcript reads derived from the (+) DNA strand. Likewise, the 5′ overhang region for each transcript derived from the (−) DNA strand was demarcated by the annotated 5′ end position of the transcript and the annotated 3′ end position of the overlapping transcript pair derived from the (+) DNA strand. Here, the 5′ overhang length was calculated by subtracting the most 3′ position of the transcript reads derived from the (+) DNA strand from the most 5′ position of the transcript reads derived from the (−) DNA strand of each complementary transcript pair.

QUANTITATION AND STATISTICAL ANALYSIS

Statistical analysis of ATP bioluminescence cell viability assays, stress granule quantitation, and quantitation of total PKR protein levels was performed with Prism version 7 (GraphPad). Statistical analysis of large datasets obtained from the Cancer Dependency Map or bulk RNA sequencing data was performed with RStudio. The statistical tests used for each displayed graph is specified in the corresponding figure legends.

Highlights.

• XRN1 is required for survival of cancer cells with interferon pathway activation

• XRN1 depletion causes activation of the interferon-inducible dsRNA sensor PKR

• Interferon signaling increases PKR levels, rendering cells vulnerable to XRN1 loss

• XRN1 loss causes accumulation of putative dsRNAs formed by sense/anti-sense RNAs