Induction of Viral Mimicry Upon Loss of DHX9 and ADAR1 in Breast Cancer Cells

Abstract Detection of viral double-stranded RNA (dsRNA) is an important component of innate immunity. However, many endogenous RNAs containing double-stranded regions can be misrecognized and activate innate immunity. The IFN-inducible ADAR1-p150 suppresses dsRNA sensing, an essential function for adenosine deaminase acting on RNA 1 (ADAR1) in many cancers, including breast. Although ADAR1-p150 has been well established in this role, the functions of the constitutively expressed ADAR1-p110 isoform are less understood. We used proximity labeling to identify putative ADAR1-p110–interacting proteins in breast cancer cell lines. Of the proteins identified, the RNA helicase DHX9 was of particular interest. Knockdown of DHX9 in ADAR1-dependent cell lines caused cell death and activation of the dsRNA sensor PKR. In ADAR1-independent cell lines, combined knockdown of DHX9 and ADAR1, but neither alone, caused activation of multiple dsRNA sensing pathways leading to a viral mimicry phenotype. Together, these results reveal an important role for DHX9 in suppressing dsRNA sensing by multiple pathways. Significance: These findings implicate DHX9 as a suppressor of dsRNA sensing. In some cell lines, loss of DHX9 alone is sufficient to cause activation of dsRNA sensing pathways, while in other cell lines DHX9 functions redundantly with ADAR1 to suppress pathway activation.


DHX9 is essential in TNBC cell lines and suppresses PKR activation 149
The similarities between ADAR1 and DHX9 led us to further study the role of DHX9 in 150 breast cancer. Analysis of publicly available data from DepMap 151 (https://depmap.org/portal/download/custom/) revealed that DHX9 is commonly essential in 152 breast cancer cell lines (Fig. 3f). We validated this finding by knocking down DHX9 in four TNBC 153 cell lines previously shown to be ADAR1-dependent 33 . In all four lines, knockdown of DHX9 154 caused cell death, likely through caspase-mediated apoptosis as indicated by cleaved PARP 155 ( Fig. 3g and h). Given the presence of the common dsRBD in DHX9 and PKR, and the role of 156 DHX9 in regulating the abundance of dsRNA 44 , we asked whether DHX9 could influence 157 activation of PKR. To our surprise, we found that in three of the TNBC cell lines studied, 158 knockdown of DHX9 caused activation of PKR ( Fig. 3g and 3i). Together these findings show 159 that DHX9 is essential in breast cancer cell lines and that in some ADAR1-dependent cell lines, 160 DHX9 suppresses PKR activation. 161

DHX9 and ADAR1 redundantly suppress PKR activation 162
The experiments above were performed in ADAR1-dependent TNBC cell lines -cell 163 lines that activate PKR following ADAR1 knockdown 33 . We were curious if DHX9 knockdown 164 would also cause PKR activation in ADAR1-independent cell lines -cell lines that do not 165 activate PKR following ADAR1 knockdown. Using shRNAs, we knocked down DHX9 in two 166 ADAR1-independent breast cancer lines, MCF-7 and SK-BR-3 ( Fig. 4a-b and 4f-g, Extended 167 Data Fig. 4a-b and 4g-h). Unlike in ADAR1-dependent cell lines, knockdown of DHX9 did not 168 cause activation of PKR in ADAR1-independent cell lines (Fig. 4a, 4d, 4f, 4i). Next, we asked 169 whether combined knockdown of ADAR1 and DHX9 in these cell lines would lead to activation 170 of PKR. As we had previously observed, knockdown of ADAR1 in SK-BR-3 and MCF-7 did not 171 cause PKR activation 33 . However, combined knockdown of DHX9 and ADAR1 caused robust 8 activation of PKR in both cell lines (Fig. 4a, 4d, 4f, 4i). Consistent with PKR activation, we 173 observed increased phosphorylation of the PKR substrate eIF2ɑ following combined knockdown 174 of ADAR1 and DHX9 ( Fig. 4a and 4d, Extended Data Fig. 4c and 4i). Like in ADAR1-dependent 175 TNBC cell lines, knockdown of DHX9 caused reduced proliferation of MCF-7 and SK-BR-3 (Fig.  176 4c, 4d, 4h, 4j), likely through caspase-dependent apoptosis, as indicated by elevated cleaved 177 PARP (Fig. 4a, 4d, Extended Data Fig. 4e, 4k). Although PARP cleavage was increased upon 178 combined knockdown of DHX9 and ADAR1, this did not statistically reduce the proliferation of 179 the cells compared to single knockdown of ADAR1 and DHX9 as measured by foci formation. 180 Together, these results reveal that ADAR1 and DHX9 redundantly suppress PKR activation in 181 ADAR1-independent breast cancer cell lines. 182

DHX9 and ADAR1 redundantly suppress RNase L activation 183
Next, we wanted to evaluate the activation of other dsRNA sensing pathways in ADAR1-184 independent cell lines after the combined knockdown of ADAR1 and DHX9. To assess whether 185 the IFN-I pathway, or other pathways, is activated after combined knockdown of ADAR1 and 186 DHX9 we turned to analysis of differential gene expression by RNA-seq. In the process of 187 preparing RNA for sequencing, we were surprised to find specific degradation of rRNA in MCF-7 188 cells following combined knockdown of ADAR1 and DHX9 (Fig. 4k). Single knockdown of either 189 DHX9 or ADAR1 did not cause rRNA degradation. The observed rRNA degradation in the 190 combined knockdown cells was reminiscent of the degradation products caused by RNase L 45 . 191 Transfection with poly(I:C), which activates the IFN-I pathway and RNase L 46 , created an 192 identical band pattern to that of combined knockdown of DHX9 and ADAR1, indicating that the 193 degradation of rRNA observed in these cells is likely caused by RNase L activity (Fig. 4k). We 194 performed the same experiment with the ADAR1-dependent TNBC cell lines described above. 195 In these cells, we did not see activation of RNase L after the knockdown of DHX9 alone 196 (Extended Data 3h).

DHX9 and ADAR1 redundantly suppress multiple innate immunity pathways 198
RNA sequencing revealed that many more RNAs were differentially expressed after 199 combined knockdown of DHX9 and ADAR1, compared to single knockdown of ADAR1 or DHX9 200 ( Fig. 5a, Extended Data Fig. 5a-f). Analysis of differential gene expression by gene set 201 enrichment after combined knockdown of ADAR1 and DHX9 in MCF-7 revealed activation of 202 multiple pathways involved in the innate response to viral infection and repression of several 203 pathways involved in translation ( Fig. 5b and Supplemental Table 14). An enrichment map 204 showed that the activated pathways associated with innate immunity formed one cluster and the 205 depressed pathways formed a separate cluster (Extended Data Fig. 5g). Of the upregulated 206 pathways, several were associated with activation of IFN signaling (Fig. 5b, and Supplementary 207 Table 14). Analysis of core ISG expression revealed significant upregulation of ISGs in MCF-7 208 after combined knockdown of ADAR1 and DHX9 ( Fig. 5c and Extended Data 5c). On the 209 contrary, knockdown of DHX9 or ADAR1 alone did not increase ISG expression. Activation of 210 RNase L described above also indicated that the type I IFN pathway was active in MCF-7 after 211 double knockdown of DHX9 and ADAR1, because the activators of RNase L, the OAS proteins, 212 are ISGs 22 . In fact, a GO term associated with OAS activity was upregulated in MCF-7 after 213 combined knockdown of ADAR1 and DHX9 (Supplementary Table 14). Consistent with 214 activation of PKR, we also observed increased expression of ATF4 targets (Fig. 5d) and NF-KB 215 targets (Fig. 5e). Interestingly, we did not observe activation of any of these pathways or RNase 216 L in SK-BR-3 following combined ADAR1 and DHX9 knockdown. 217 The finding that combined knockdown of ADAR1 and DHX9 did not induce ISG 218 expression in SK-BR-3 is consistent with our findings in TNBC cell lines. While knockdown of 219 DHX9 alone caused activation of PKR in several TNBC cell lines, we observed no activation of 220 the type I IFN pathway, as indicated by no change in ISG15 expression (Extended Data Fig. 3f-221 g). ISG15 was found to be highly upregulated at the RNA and protein level in MCF-7 after double knockdown of ADAR1 and DHX9 (Extended Data Fig. 4f, 4m, Supplementary Table 8).  223 Like ISG expression overall, ISG15 expression in SK-BR-3 was not changed by knockdown of 224 DHX9 and/or ADAR1 (Extended Data Fig. 4l, 4n). 225 Given the previously described role of DHX9 in the control of Alu containing RNAs, we 226 next sought to assess if increased expression of transposable elements, especially Alus, could 227 explain the activation of PKR or the IFN pathway upon combined knockdown of DHX9 and 228 ADAR1. Analysis of our RNA-seq data revealed that transposable element expression was 229 generally unchanged upon either single knockdown of ADAR1 or DHX9, or combined 230 knockdown of both proteins (Extended Data Fig. 6a-f). 231 The dsRBDs of DHX9 are sufficient to suppress PKR activation in the absence of ADAR1 232 Having shown through knockdown studies that ADAR1 and DHX9 function redundantly 233 to suppress dsRNA sensing, we next wanted to assess which functions of DHX9 and which 234 isoforms of ADAR1 are important for this role. To determine which functions of DHX9 are 235 sufficient to suppress PKR activation, we performed a rescue experiment with wild-type DHX9, 236 a helicase deficient mutant DHX9, and a truncated DHX9 that possess the N-terminal dsRBDs 237 fused to EGFP (Fig. 6a). Overexpression of wild-type DHX9 rescued the PKR activation 238 phenotype caused by double knockdown of ADAR1 and DHX9, confirming that the observed 239 phenotypes are not an off-target effect of the shRNAs used for knockdown ( Fig. 6b and 6d). 240 Interestingly, the DHX9 K417R mutant, which lacks helicase activity due to its inability to bind 241 ATP 47 , was also capable of suppressing PKR activation. The same was true for a construct 242 which contained the dsRBDs of DHX9 fused to EGFP (dsRBD-EGFP). These findings indicate 243 that the DHX9 dsRBDs are likely sufficient to suppress PKR activation in the absence of ADAR1 244 and DHX9. However, only wild-type DHX9 could rescue the reduced foci formation observed after DHX9 knockdown ( Fig. 6c and 6e). This finding indicates that the PKR activation 246 phenotype and the reduced proliferation phenotype are uncoupled. 247 Consistent with the observation that the DHX9 dsRBDs are sufficient to suppress PKR 248 activation, knockdown of DDX17, which lacks dsRBDs, did not cause substantial PKR activation 249 in SK-BR-3 (Extended Data Fig. 9). Unlike DHX9, knockdown of DDX17 had no effect on cell 250 proliferation as measured by the foci formation assay (Extended Data Fig. 9b and 9d). While 251 combined knockdown of DHX9 and ADAR1 in SK-BR-3 caused a 5-10 fold increase in PKR 252 phosphorylation, the combined knockdown of DDX17 and ADAR1 caused only a modest 1.5-253 fold increase in PKR phosphorylation (Extended Data Fig. 9a and 9c). This finding underscores 254 the novel role of the DHX9 helicase and its dsRBD, a unique domain among this large family of 255

RNA helicases. 256
The p110 and p150 isoforms of ADAR1 suppress PKR activation in the absence of DHX9 257 Next, we turned to ADAR1, and asked which isoform of ADAR1 is sufficient to suppress 258 PKR activation in the absence of DHX9. We used the same approach as above, a rescue 259 experiment with overexpression of ADAR1-p110 or ADAR1-p150. Interestingly, we found that 260 both ADAR1 isoforms were sufficient to suppress PKR activation upon loss of DHX9 ( Fig. 6f and 261 6h, Extended Data Fig. 8a-b). However, overexpression of neither ADAR1 isoform was able to 262 rescue the foci formation phenotype, again indicating that the PKR activation and cell 263 proliferation phenotypes are uncoupled ( Fig. 6g and 6i). 264

Discussion 265
In recent years, ADAR1 has become an important therapeutic target for breast and other 266 cancers. It is clear from the literature that depletion of ADAR1 in ADAR1-dependent cell lines 267 leads to activation of dsRNA sensors and innate immunity programs that lead to cell 268 death 21,32,33 . Yet unclear is what distinguishes ADAR1-dependent cell lines from ADAR1-independent cell lines -those that are insensitive to ADAR1 depletion. Elevated ISG expression 270 has been proposed as a potential prerequisite for ADAR1-dependency, but some ADAR1-271 independent cell lines exhibit elevated ISG expression 21,33 . As such, more information is needed 272 to identify the factors that establish ADAR1-dependency or ADAR1-independency. To begin to 273 fill in some of the knowledge gaps surrounding ADAR1, we utilized proximity labeling to identify 274 putative ADAR1 interacting proteins, specifically focusing on the less studied ADAR1-p110 275 isoform. 276 Of the proteins identified by proximity labeling, the DEAH box helicase DHX9 was of 277 particular interest. Like ADAR1 and PKR, DHX9 possesses dsRBDs, a singularly unique trait 278 among the DEAD/DEAH box RNA helicase family. DHX9 expression is strongly correlated with 279 ADAR1-p110 expression in breast cancer. Interestingly, both genes are located on the q-arm of 280 chromosome 1, though they are separated by 28 mb and are thus unlikely to be physically co-281 regulated. Consistent with other reports in the literature, we show here that ADAR1 and DHX9 282 likely interact directly 38 . We observed that ADAR1-p110 and DHX9 interact in an RNA-283 independent manner, while the interaction between ADAR1-p150 and DHX9 was disrupted by 284 RNase treatment. This result contradicts previous findings, in which ADAR1-p150 and DHX9 co-285 immunoprecipitated after RNase A treatment 44 . This discrepancy may be due to different cell 286 lines used, HEK293T was used for the previous study, or different accessibility of DHX9 287 epitopes during immunoprecipitation. The importance of the ADAR1-DHX9 interaction is unclear 288 from this work. Further studies are needed to structurally assess the interaction and directly 289 perturb the interaction to understand what function it may have. 290 Here we report that in addition to being a commonly essential gene in breast cancer, 291 DHX9 suppresses dsRNA sensing. In ADAR1-dependent cell lines, knockdown of DHX9 alone -292 much like knockdown of ADAR1 alone -caused activation of the dsRNA sensor PKR 33 . Like 293 ADAR1 knockdown, DHX9 knockdown had no effect on PKR activation in ADAR1-independent 294 cell lines. However, combined knockdown of DHX9 and ADAR1 caused robust activation of 295 Cottrell, Ryu et al., 2023 13 PKR in these cells. This finding indicates that ADAR1 and DHX9 function redundantly to 296 suppress PKR activation in ADAR1-independent cell lines and provides an explanation for why 297 PKR is not activated in ADAR1-independent cells upon ADAR1 knockdown. Interestingly, like 298 ADAR1, DHX9 has been shown to interact with PKR and is phosphorylated by PKR 48 . More 299 research is needed to understand the importance of PKR phosphorylation of DHX9 and whether 300 or not it may serve as a feedback mechanism. 301 Rescue experiments revealed that the helicase activity of DHX9 was dispensable for 302 suppression of PKR activation. In fact, the N-terminal dsRBDs of DHX9 were sufficient to 303 suppress PKR activation in the absence of ADAR1 in ADAR1-independent cell lines. This 304 finding provides evidence for a model in which DHX9 competes with PKR for dsRNA binding 305 through its dsRBDs (Fig. 7). This competition is likely to be indirect, as DHX9 is nuclear 306 localized while PKR is generally localized in the cytosol 49-52 . As such, DHX9 may function to 307 suppress PKR activation by sequestering dsRNAs in the nucleus. 308 Previously, the ADAR1 p150 isoform, and not the p110 isoform, was shown to be 309 responsible for suppression of PKR activation. Through rescue experiments, we show here that 310 both isoforms are sufficient to suppress PKR activation in the absence of DHX9 in ADAR1-311 independent cell lines. A preprint that was published during the preparation of this manuscript 312 showed that the dsRBDs of ADAR1, ADAR2, and STAU1 were sufficient to suppress PKR 313 activation 53 . Based on this finding and our rescue experiment with the DHX9 dsRBDs, it is likely 314 that ADAR1-p150 and ADAR1-p110 suppress dsRNA sensing in the absence of DHX9 by 315 competing with PKR for dsRNA binding. Although this may be direct competition in the case of 316 ADAR1-p150 localized cytoplasmically, ADAR1-p110 may function like DHX9 to sequester 317 dsRNAs in the nucleus. 318 In addition to suppression of PKR activation, we also observed that DHX9 and ADAR1 319 redundantly suppress activation of several other dsRNA sensing pathways in MCF-7. 320 Knockdown of both proteins caused activation of IFN-I signaling, likely via MDA5 activation, as previously shown for ADAR1 13,15 . We also observed activation of OAS-RNaseL and increased 322 expression of the ATF4 and NF-KB targets, likely downstream of PKR activation 54-56 . Taken 323 together, combined knockdown of DHX9 and ADAR1 in MCF-7 creates a viral mimicry 324 phenotype, where multiple innate immune pathways against RNA viruses have been activated 325 (Fig. 7). Interestingly, we only observed activation of PKR in other cell lines after either DHX9 326 knockdown alone or combined knockdown with ADAR1. Two possible explanations for this 327 discrepancy could be 1) Another factor may be present in some cells that suppresses dsRNA 328 sensing in the absence of DHX9 and/or ADAR1. 2) The expression of endogenous dsRNAs that 329 cause activation of sensors other than PKR may vary, leading to the differential effects of 330 ADAR1 and/or DHX9 knockdown. Further studies are needed to identify the immunogenic 331 RNAs that activate the various sensors. 332 Induction of viral mimicry has great potential as a therapeutic approach for multiple 333 cancers, including TNBC 57-60 . In addition to the cell intrinsic effects of activating innate immune 334 pathways within the tumor, the signaling that occurs after activation of those pathways can 335 promote anti-tumor immunity, especially when combined with checkpoint inhibitors 61-63 . 336 Combined therapies targeting ADAR1 and DHX9 may serve as an effective means of treating 337 breast and other cancers by inducing viral mimicry. 338 The sequences for shRNAs against DHX9 and DDX17 are in Supplementary Table 1.  Table 1). For shADAR1, the shRNA was subcloned into pLKO. Streptavidin magnetic beads were washed twice with RIPA containing HALT and quenching 405 agents. The lysate from above was incubated with the beads for 1 hour at room temperature. 406

Materials and Methods
The beads were washed in the following order: once with RIPA containing HALT and quenching 407 agents, once with RIPA, once with 1 M KCl, once with 2 M urea pH 8.0, twice in RIPA and once 408 with water. Elution was performed in 1X SDS-Sample buffer by heating at 95 °C for 10 minutes. 409 The eluate was analyzed by LC-MS-MS, see below. 410

Mass Spectrometry 411
Liquid-chromatography and tandem mass spectrometry was performed by MSBioWorks 412 (Ann Arbor, MI). The eluates from the streptavidin pulldown above were processed by SDS-413 PAGE using 10% Bis-Tris NuPage Mini-gel (Invitrogen) with the MES buffer system. The gel 414 was run 2cm. The mobility region was excised and processed by in-gel digestion with trypsin 415 using a robot (ProGest, DigiLab). For the trypsin digestion, the gel slices were washed with 416 25mM ammonium bicarbonate followed by acetonitrile. The samples were reduced with 10mM dithiothreitol at 60°C followed by alkylation with 50mM iodoacetamide at room temperature. 418 Subsequently proteins were digested with trypsin (Promega) at 37°C for 4 h. The trypsin 419 digestion was quenched with formic acid and the supernatant was analyzed directly without 420 further processing. The were washed briefly with 1x PBS prior to fixation in 100% methanol for 5 min. After drying, the 515 cells were stained with 0.005% Crystal Violet solution containing 25% methanol (Sigma-Aldrich) 516 prior to washing excess stain away with deionized water. The plates were scanned using an 517 ImageScanner III (General Electric). Foci area was calculated using ImageJ. 518

Analysis of CCLE and TCGA data 519
RNA-seq normalization and calculation of z-scores was performed as described 520