Chronic viral mimicry induction following p53 loss promotes immune evasion

Abstract

Epigenetic therapies facilitate transcription of immunogenic repetitive elements that cull cancer cells through ‘viral mimicry’ responses. Paradoxically, cancer-initiating events also facilitate transcription of repetitive elements. Contributions of repetitive element transcription towards cancer initiation, and the mechanisms by which cancer cells evade lethal viral mimicry responses during tumor initiation remain poorly understood. In this report, we characterize premalignant lesions of the fallopian tube along with syngeneic epithelial ovarian cancer models to explore the earliest events of tumorigenesis following loss of the p53 tumor suppressor protein. We report that p53 loss permits transcription of immunogenic repetitive elements and chronic viral mimicry activation that increases cellular tolerance of cytosolic nucleic acids and diminishes cellular immunogenicity. This selection process can be partially attenuated pharmacologically. Altogether, these results reveal that viral mimicry conditioning following p53 loss promotes immune evasion and may represent a pharmacological target for early cancer interception

INTRODUCTION

Repetitive elements within the human genome are subject to transcriptional silencing by DNA- and histone modifications to prevent the deleterious consequences associated with their transcriptional reactivation. Emerging epigenetic therapies confer anticancer effects, in part, through disruption of such DNA- and histone modifications to permit acute derepression of repetitive elements genome wide. A subset of derepressed repetitive elements form cytosolic DNAs and dsRNAs that stimulate cytosolic DNA- and RNA- sensors respectively to activate an interferon-dependent antiviral response in the absence of exogenous viral infection, a phenomenon coined ‘viral mimicry’. The discovery of viral mimicry as a negative selection pressure to cull cancer cells is difficult to reconcile with the paradoxical observation that nearly every human cancer exhibits aberrant transcriptional reactivation of repetitive elements. Whether cancers induce or evade viral mimicry following transcriptional reactivation of repetitive elements remains poorly understood. An emerging body of literature implicates cancer-initiating events with perturbed epigenetic silencing of repetitive elements(1). Therefore, investigating mechanisms of tumor initiation harbors the potential to reveal possible contributions of viral mimicry during tumorigenesis.

Amongst the cancer-initiating events implicated in transcriptional reactivation of repetitive elements, functional inactivation of canonical tumor suppressor proteins features prevalently. The most frequently mutated tumor suppressor gene in human cancer is TP53, which encodes the p53 tumor suppressor protein(2). Ovarian high-grade serous carcinomas (HGSCs) in particular exhibit nearly 100% penetrance of somatic TP53 mutations as a cancer-initiating event(3). HGSCs in humans predominantly originate in the fallopian tube from TP53 mutant cells, referred to as p53 signatures. Patients may harbor p53 signatures or serous tubal intraepithelial lesions (STILs) for 20–25 years before onset of premalignant serous tubal intraepithelial carcinomas (STICs) that persist for several years prior to HGSC. Notably, only a subset of STICs progress to HGSC, for reasons that remain unclear(4). In mice, HGSC can arise from either ovarian epithelium or oviduct following identical genetic alterations of several tumor suppressor genes that include the TP53 murine orthologue Trp53(5). Indeed, all murine HGSCs require Trp53 mutations.

Curiously, TP53 mutations are required but insufficient to promote transformation(6–9). Cancer-associated TP53 mutations are detectable in epithelial tissues of healthy individuals, including gynecological tissues(10). The additional events required to progress from a healthy cell to premalignant lesion and finally a malignant tumor remain poorly understood. Importantly, the acquisition of additional recurrent mutations is not required as premalignant STICs progress to HGSCs(4). Moreover, the possible involvement of repetitive elements or viral mimicry during HGSC initiation remains unclear.

In this report, we perform whole transcriptome profiling of STICs and adjacent normal fallopian tissue. We discover that STICs are distinguished from adjacent normal tissue by transcriptional dysregulation of repetitive elements. Cellular models reveal that p53 inactivation disrupts constitutive heterochromatinization of repetitive elements. Subsequently expressed retrotransposons stimulate cytosolic RNA- and DNA sensors to initiate the viral mimicry response. In contrast to acute viral mimicry, constitutive agonism of DNA- and RNA sensors selects for diminished IFN induction in concert with increased tolerance for elevated cytosolic nucleic acids. This selection process is characterized by diminished p300/CBP-dependent H3K27Ac at promoters of interferon stimulated genes (ISGs), reduced ISG expression, and reduced secretion of proinflammatory cytokines. Accordingly, viral mimicry ‘tolerant’ cells exhibit alterations to the tumor infiltrating lymphocyte milieu, diminished antitumor adaptive immune responses, and diminished response to viral mimicry-inducing epigenetic therapy. This immune evasion mechanism that we term ‘viral mimicry conditioning’ can be partially attenuated using the FDA-approved reverse transcriptase inhibitor lamivudine (3TC). The results presented here provide context for retrotransposon dysregulation that initiates in premalignancy and persists throughout tumorigenesis, and frames viral mimicry conditioning as a potential target for early cancer interception.

RESULTS

Premalignant lesions exhibit cell cycle dysregulation, inflammation, and p53-sensitive retrotransposon dysregulation

Nearly all ovarian high-grade serous carcinomas (HGSCs) initiate with acquisition of somatic TP53 mutations in fallopian tube secretory epithelial cells (FTSECs)(11). The subsequent transcriptional changes that characterize premalignancy and early tumorigenesis remain poorly understood. Therefore, we conducted laser capture microdissection (LCM) to isolate serous tubal intraepithelial carcinomas (STICs) and normal fallopian tube epithelium from the same patient on FFPE blocks for RNA-seq (Figure S1A, Figure 1A–B). Cases from germline BRCA1/2 wildtype patients were selected to specifically assess the cancer-initiating contributions of somatic TP53 mutations in isolation (Table S1). Immunohistochemistry reveals expected elevations in Ki67 in STIC lesions, along with elevated p53 staining characteristic of missense TP53 mutations (Figure 1A). Our cohort also includes STICs that exhibit p53 null staining patterns characteristic of nonsense TP53 mutations (Figure 1A).

Figure 1: TP53 mutation facilitates transcriptional dysregulation of retrotransposons in STICs and ovarian epithelial cells.

(A) Bright field images of H&E and IHC of STIC sections. (B) Bright field images of pre vs. post-laser capture microdissection (LCM) of FFPE STIC lesions (C) Enriched terms in STIC lesions relative to adjacent normal determined from Gene Ontology (GO) analysis of RNA-seq performed on tissues from eight patients. (D) Mean average (MA) plots of all repetitive element expression changes (left) and for intergenic repetitive elements (right) in dissected STIC lesions relative to normal adjacent tissue determined by RNA-seq. Upregulated and downregulated repeats are depicted with red and blue dots respectively, while gray dots represent repeats that are not significantly changed. Y axis represents log2FC, and x axis represents Average log2CPM. Statistical significance of up/downregulated repeat elements was determined by|log₂ FC|≥1 and FDR<0.05. (E) Donut plot displays proportions of upregulated repeat classes (n=12,260) in dissected STIC lesions relative to normal adjacent tissue determined by RNA-seq. Counts of upregulated repeat elements of each class were compared with counts of whole genome repeats of the corresponding to calculate odds ratio and p-values using the two-sided Fisher exact test. Significance level is “*”:p<0.05. “**”:p<0.01, “***”:p<0.001, and “****”:p<0.0001. (F) GSEA pathway analysis showing the IFN⍺ pathway in STIC lesions relative to normal adjacent tissue determined by RNA-seq. (G) MA plot of all repeat elements (left) and for only intergenic repeat elements (right) expression changes in TP53^−/− fallopian tube secretory epithelial cells (FTSECs) relative to TP53^+/+ FTSECs determined from RNA-seq. Upregulated and downregulated repeats are depicted with red and blue dots respectively, while gray dots represent repeats that are not significantly changed. Y axis represents log2FC, and x axis represents Average log2CPM. Statistical significance of up/downregulated repeat elements was determined by|log₂ FC|≥1 and FDR<0.05. (H) Donut plot displays the proportions of upregulated repeat classes (n=163) in TP53^−/− relative to TP53^+/+ in FTSEC samples. Counts of upregulated repeat elements of each class were compared with counts of whole genome repeats of the corresponding class to calculate odds ratio and p-values using the two-sided Fisher exact test. Significance level is “*”:p<0.05, “**”:p<0.01, “***”:p<0.001, and “****”:p<0.0001. (I) GSEA analysis showing the enrichment of IFN⍺ pathway in TP53^−/− relative to TP53^+/+ in FTSEC samples. (J) MA plot of all repeat elements (left) and for only intergenic repeat elements (right) expression changes in Trp53^−/− ID8 cells relative to Trp53^+/+ ID8 cells determined from RNA-seq. Upregulated and downregulated repeats are depicted with red and blue dots respectively, while gray dots represent repeats that are not significantly changed. Y axis represents log2FC, and x axis represents Average log2CPM. Statistical significance of up/downregulated repeat elements was determined by|log₂ FC|≥1 and FDR<0.05. (K) Donut plot displays the proportions of upregulated repeat classes (n=3,101) in Trp53^−/− relative to Trp53^+/+ in ID8 cells. Counts of upregulated repeat elements of each class were compared with counts of whole genome repeats of the corresponding class to calculate odds ratio and p-values using the two-sided Fisher exact test. Significance level is “*”:p<0.05, “**”:p<0.01, “***”:p<0.001, and “****”:p<0.0001. (L) GSEA analysis showing the enrichment of IFN⍺ pathway in Trp53^−/− relative to Trp53^+/+ in ID8 cells. (M) MA plot of all repeat elements (left) and for only intergenic repeat elements (right) expression changes in Trp53^−/− oviduct OVE4 relative to Trp53^+/+ oviduct OVE4 determined from RNA-seq. Upregulated and downregulated repeats are depicted with red and blue dots respectively, while gray dots represent repeats that are not significantly changed. Y axis represents log2FC, and x axis represents Average log2CPM. Statistical significance of up/downregulated repeat elements was determined by|log₂ FC|≥1 and FDR<0.05. (N) Donut plot displays the proportions of upregulated repeat classes (n=2,292) in Trp53^−/− relative to Trp53^+/+ in oviduct OVE4. Counts of upregulated repeat elements of each class were compared with counts of whole genome repeats of the corresponding class to calculate odds ratio and p-values using the two-sided Fisher exact test. Significance level is “*”:p<0.05, “**”:p<0.01, “***”:p<0.001, and “****”:p<0.0001. (O) GSEA analysis showing the enrichment of IFN⍺ pathway in Trp53^−/− relative to Trp53^+/+ in oviduct OVE4. (P) MA plot of all repeat elements (left) and for only intergenic repeat elements (right) expression changes in Trp53^−/− oviduct OVE16 relative to Trp53^+/+ oviduct OVE16 determined from RNA-seq. Upregulated and downregulated repeats are depicted with red and blue dots respectively, while gray dots represent repeats that are not significantly changed. Y axis represents log2FC, and x axis represents Average log2CPM. Statistical significance of up/downregulated repeat elements was determined by|log₂ FC|≥1 and FDR<0.05. (Q) Donut plot displays the proportions of upregulated repeat classes (n=423) in Trp53^−/− relative to Trp53^+/+ in oviduct OVE16. Counts of upregulated repeat elements of each class were compared with counts of whole genome repeats of the corresponding class to calculate odds ratio and p-values using the two-sided Fisher exact test. Significance level is “*”:p<0.05, “**”:p<0.01, “***”:p<0.001, and “****”:p<0.0001. (R) GSEA analysis showing the enrichment of IFN⍺ pathway in Trp53^−/− relative to Trp53^+/+ in oviduct OVE16.

Consistent with abrogation of canonical p53 functions, the majority of upregulated gene sets in STIC lesions relative to adjacent normal tissue correspond to genes involved in cell cycle entry and progression (Figure 1C, Figure S1B–C). For example, STIC lesions exhibit pronounced upregulation of cyclins and cyclin-dependent kinases (CDK4, CDK2, CDK1, CCNE1, CCNA2, CCNB2), along with targets involved in DNA replication (E2F1–3, MCM2,3,4,6,7, ORC6, CDC20, CDC7, MYBL2) and mitotic progression (AURKA, MAD2L1, BUB1). p53-activated genes, such as CDKN1A that encodes p21, exhibit expected downregulation (Figure S1B).

We next sought to determine whether the repetitive element dysregulation associated with abrogated p53 function observed in other biological contexts(12) could be observed in STIC lesions. Indeed, RNA-seq reveals pronounced upregulation and downregulation of repetitive elements in STIC lesions relative to adjacent normal epithelium (Figure 1D; Figure S1D–H). Detection of intergenic repeat expression reveals that STICs express both autonomously transcribed- and co-transcribed repetitive elements embedded in introns, coding sequences, and UTRs (Figure 1D; Figure S1D). The majority of upregulated repetitive elements belong to retrotransposon classes of repetitive elements that predominantly reside in intronic- or intergenic regions (Figure 1E; Figure S1D). Importantly, autonomously transcribed- and co-transcribed retrotransposons are capable of forming immunogenic nucleic acids capable of stimulating viral mimicry responses(13). Therefore, we assessed potential transcriptional signatures indicative of viral mimicry induction. Surprisingly, Gene Set Enrichment Analysis (GSEA) reveals lack of Type I interferon (IFN) induction, inconsistent with a pronounced viral mimicry response that might be expected with prevalent retrotransposon upregulation (Figure 1F: Figure S1C).

We sought to explore the cause of early repeat misexpression during HGSC pre-malignancy, and whether this pattern of elevated retrotransposon expression coupled with dampened viral mimicry induction occurs in cellular models of HGSC initiation. We assessed whether p53 loss perturbs transcriptional regulation of repetitive elements in human fallopian tube secretory epithelial cells. RNA-seq from isogenic TP53^+/+ and TP53^−/− FTSECs(14) reveals that p53 loss alters expression of elements from all major repeat classes in both directions (Figure 1G; Figure S1I–M). Specifically, p53 loss upregulates LTR, LINE, and SINE retrotransposons that predominantly reside within intronic and intergenic regions (Figure 1H, Figure S1I). Despite elevated retrotransposon expression, p53-deficient FTSECs exhibit a predominantly negative enrichment score associated with Type I interferon (IFN) induction (Figure 1I).

Murine ovarian epithelial cancers can be generated from identical genetic perturbations in either oviduct epithelial cells or ovarian surface epithelial cells(5). Therefore, we assessed whether the relationship between p53 function and repeat regulation observed in human HGSC premalignant lesions and cell-of-origin models could be observed in murine cells. Indeed, RNA-seq reveals immortalized Trp53^−/− ID8 ovarian surface epithelial cells exhibit pronounced upregulation and downregulation of co-transcribed- and autonomously transcribed repetitive sequences relative to isogenic Trp53^+/+ cells (Figure 1J; Figure S2A–E). Despite upregulated intronic and intergenic LTR, LINE, and SINE retrotransposons (Figure 1K, Figure S2A), Trp53^−/− ID8 cells exhibit a negative enrichment score associated with Type I IFN (Figure 1L) that include genes involved in the viral mimicry response (Figure S2F–I). This pattern of retrotransposon upregulation coupled with a negative enrichment score associated with Type I interferon (IFN) induction in p53-deficient cells was recapitulated in murine oviduct OVE4 (Figure 1M–O, Figure S2J–N) and OVE16 (Figure 1P–R, Figure S2O–S) cells. Notably, more pronounced transposable element expression changes in STIC lesions compared to isogenic cell lines with p53 loss may reflect additional epigenetic aberrations that accrue during premalignancy.

Collectively, RNA-seq from multiple HGSC cell-of-origin models reveals that pronounced transcriptional dysregulation of repetitive elements observed in STIC lesions can be recapitulated following p53 loss in human FTSECs, murine ovarian surface epithelial cells, and murine oviductal epithelial cells. Paradoxically, upregulation of retrotransposons is not associated with pronounced induction of a viral mimicry response in STIC lesions or cellular models of HGSC initiation.

p53 loss diminishes H3K9me3 at repetitive DNA elements

We sought to explore mechanistic connections between p53 loss and transcriptional dysregulation of repetitive elements. p53 is best characterized as a transcription factor that occupies target gene promotors under conditions of cellular stress such as DNA damage(2). Surprisingly, CUT&RUN reveals extensive p53 chromatin occupancy in ovarian epithelial cells during normal homeostasis (Figure 2A). Gene Ontology annotation of CUT&RUN peaks indicates p53 occupies bona fide p53 target genes involved in DNA damage responses(15), such as Cdkn1a and Bbc3, under normal homeostasis (Figure 2B–C; Figure S3A–B). Accordingly, p53 CUT&RUN peaks exhibit enrichment for known p53 recognition motifs associated with occupancy under DNA damaging conditions (Figure S3C). Therefore, p53 occupies promoters of canonical p53 target genes involved in DNA damage responses during normal homeostasis in ovarian epithelial cells.

Figure 2: p53 loss diminishes H3K9me3 at transcriptionally upregulated repetitive elements.

(A) Heatmap and average profile of p53 CUT&RUN signal in Trp53^+/+ and Trp53^−/− ID8 cells at wildtype-specific p53 peaks (n=2,127). Signal is plotted at peak center +/− 2Kb. (B) Genome tracks depict CUT&RUN signals at Cdkn1a (left) and Bbc3 (right) for p53 in Trp53^+/+ and Trp53^−/− ID8 cells, and IgG in Trp53^+/+ ID8 cells under non-treated conditions. (C) Bar plot depicts −log10(p-value) of enriched pathways from Gene Ontology analysis of wildtype-specific p53 CUT&RUN peaks (n=2,127) performed using GREAT analysis tool. p-value was calculated using the binomial test. (D) Donut plot depicts genomic distribution of wildtype-specific p53 CUT&RUN peaks (n=2,127). Counts of peaks at each genomic region were compared with the shuffled version of the same peaks to calculate the odds ratio and p-value using the two-sided Fisher exact test. Significance level is “*”:p<0.05, “**”:p<0.01 and “****”:p<0.0001. (E) Donut plot depicts wildtype-specific p53 CUT&RUN peaks (n=2,127) distribution per repeat classes. Counts of peaks were compared with the counts of repeat classes in the whole genome to calculate the odds ratio and p-value using two-sided Fisher exact test. Significance level is “*”:p<0.05, “**”:p<0.01, “***”:p<0.001, and “****”:p<0.0001. (F) Genome track plot depicts p53 CUT&RUN signal in Trp53^+/+ and Trp53^−/− ID8 cells and IgG signal in Trp53^+/+ genotype for a locus that includes a wildtype p53 CUT&RUN peak overlapping with an intergenic RLTR1B repeat element. (G) Heatmaps and average profiles of p53 CUT&RUN signal in Trp53^+/+ and Trp53^−/− ID8 cells for repeat elements that overlap with wildtype-specific p53 CUT&RUN peaks stratified by repeat class. Signal tracks are plotted from the 5’end to the 3’end of the repeat element +/− 0.5Kb. Heatmaps from left to right correspond to LTR (n=1,062), LINE (n=105), and SINE (n=263) repeat classes. (H) Heatmaps and average profiles of p53 CUT&RUN signal in Trp53^+/+ and Trp53^−/− ID8 cells for repeat elements from selected LTR subfamilies that overlap with wildtype-specific p53 CUT&RUN peaks. Signal tracks are plotted from the 5’end to the 3’end of the repeat element +/− 0.5Kb. Heatmaps from left to right correspond to RLTR1B/ERV1 (n=419), MMERGLN_LTR/ERV1 (n=276), RLTR23/ERV1 (n=103), RLTR1F_Mm/ERVK (n=36), and MTD/ERVL-MaLR (n=14). (I) Heatmaps and average profiles of H3K9me3 CUT&RUN signal in Trp53^+/+ and Trp53^−/− ID8 cells at repeats upregulated in Trp53^−/− ID8 cells relative to Trp53^+/+ ID8 cells as determined from RNA-seq. Signal tracks are plotted from the 5’end to the 3’end of repeat elements +/− 2Kb. Heatmaps from left to right correspond to LTR (n=947), LINE (n=198), SINE (n=1,835), and DNA transposon (n=92) repeat classes. Statistical significance of upregulated repeat elements was determined by log₂ FC ≥ 1 and FDR<0.05. (J) Genome tracks depict the RNA-seq and CUT&RUN signals of H3K27me3 and H3K9me3 at two RLTR6_Mm repeat loci.

~15% of p53 peaks map to gene promoters (Figure 2D, S3B). Instead, the majority of p53 peaks map to repeat-rich intergenic and intronic regions represented by all major repeat classes (Figure 2E–G; Figure S3D–E). In particular, LTR families of endogenous retroviruses that include ERV1, ERVK, and ERVL families exhibit pronounced 5’ enrichment of p53 (Figure 2F–H, Figure S3E–F). p53-bound ERV1 elements exhibit enrichment of known p53-specific motifs (Figure S3G). Therefore, p53 predominantly occupies LTRs that harbor known p53 recognition motifs.

Surprisingly, repetitive elements that increase in expression upon p53 loss exhibit minimal p53 occupancy in wildtype cells (Figure S3H). This suggests that p53-sensitive alterations to repeat transcription may occur through indirect effects on repeat loci. p53 exerts transcriptional regulation through an extensive interactome that includes chromatin regulatory proteins(2). Therefore, we used CUT&RUN to assess broad changes in heterochromatin distribution at repetitive elements. We observe that H3K9me3 is reduced at repetitive elements that are transcriptionally upregulated in p53 deficient ID8 cells (Figure 2I–J, Figure S3I), suggesting a lack of compensatory mechanisms to maintain transcriptional repression of these elements following p53 loss. Importantly, total H3K9me3 levels appear equivalent between Trp53^+/+ and Trp53^−/− ID8 cells (Figure S3J), and expression of H3K9 histone methyltransferases (HMTs) and H3K9me3 demethylases remain consistent across genotypes (Figure S3K). Furthermore, H3K27me3 does not exhibit p53-dependent changes at repetitive DNA sequences in ID8 cells (Figure S3L). Finally, H3K4me3 levels remain relatively equivalent between Trp53^+/+ and Trp53^−/− ID8 cells at repetitive elements upregulated in p53-deficient cells (Figure S3M–N). Collectively, these observations suggests that p53 loss diminishes H3K9me3 at transcriptionally dysregulated repetitive elements.

p53 loss elevates cytosolic RNA:DNA and dsRNA levels

To understand why dysregulation of repetitive elements following p53 loss does not induce pronounced viral mimicry induction in our experimental models, we assessed transcriptional activation of repetitive elements that are capable of forming immunogenic agonists of cytosolic DNA- or RNA-sensors. LINEs can form immunogenic reverse transcription products(16) that induce chronic sterile inflammation in aging murine tissues through agonism of cytosolic DNA sensors(17). While LINE reverse transcription occurs in the nucleus, reverse-transcription products may accumulate in the cytosol as observed in macular degeneration(18,19) and RNaseH-deficient Aicardi–Goutières syndrome(20). The recent discovery of cytosolic priming of the L1 reverse transcriptase might provide one potential mechanistic explanation of such observations(16). Diminished IFN signaling following pharmacological inhibition of LINE reverse transcriptase activity suggests that cytosolic reverse-transcription products are immunostimulatory(16,19,21).

We assessed expression of the L1 ORF1p RNA binding protein(22) to determine if STICs express machinery required to generate reverse transcription intermediates. Strikingly, immunohistochemistry reveals pronounced expression of L1 ORF1p specifically in cells that exhibit a positive p53 IHC signal indicative of mutant p53 expression. Notably, L1 ORF1p is not expressed in adjacent normal fallopian tube epithelium (Figure 3A). STICs that exhibit a p53 null staining pattern, indicative of a TP53 nonsense mutation, express L1 ORF1p specifically in the precursor lesion but not the adjacent normal tissues (Figure 3A–B). Our results corroborate earlier reports that note the same associations in STICs from patients with germline BRCA1/2 mutations or unknown BRCA1/2 status(23,24). Collectively, pronounced L1 ORF1p staining in BRCA1/2 wildtype STICs suggests that missense or nonsense TP53 mutations are sufficient to upregulate L1 ORF1p in fallopian tube epithelial cells independent of BRCA1/2 mutations implicated in retrotransposon regulation in other contexts(25).

Figure 3: p53 loss increases cytosolic accumulation of RNA:DNA and dsRNA.

(A) Bright field images of IHC performed on STIC sections for p53 and L1 ORF1p. p53 IHC image of STIC004 is lower magnification of corresponding panel from Figure 1A. (B) Pathologist scoring of L1 ORF1p staining strength and staining pattern per STIC lesion (C) Donut plot depicts proportions of LINE families upregulated in STIC lesions versus adjacent normal tissue determined from RNA-seq. Counts of repeat elements were compared with counts of whole genome repeat element to calculate odds ratio and p-values using the two-sided Fisher exact test. Significance level is “*”:p<0.05, “**”:p<0.01, “***”:p<0.001, and “****”:p<0.0001. (D) Distribution of L1 element length within human genome versus those upregulated in STIC lesions as detected by RNA-seq. (E-F) Genome tracks depict the RNA-seq signal of two upregulated L1 loci in STICs relative to adjacent normal tissue from 8 patients. (G) qPCR from RNA strand of cytoplasmic RNA:DNA from ID8 cells. (H) Scatter plot showing the annotated IR Alus and MDA5-binding IR Alus that are upregulated in STICs relative to adjacent normal tissue from 8 patients. Red and purple dots represent upregulated annotated IR Alus and MDA5-binding IR Alus respectively, and blue and green dots represent downregulated annotated IR Alus and MDA5-binding IR Alus respectively. Gray dots represent other IR Alus that are not significantly regulated. The number of upregulated MDA5-binding IR pairs was compared with the number of annotated IR Alu pairs using the Fisher exact test to calculate the p-value and odds ratio. (I) Genome tracks depict the RNA-seq signal of an IR Alu pair upregulated in STICs relative to adjacent normal tissue from 8 patients. (J) Scatter plot showing the IR Alus that overlap with dsRNA regions of high force that are upregulated in STICs relative to adjacent normal tissue in 8 patients. Red and blue dots represent upregulated and downregulated IR Alus with high force respectively, while gray dots represent other high force IR Alus that are not significantly upregulated. The count of upregulated IR Alus was compared with the count of downregulated IR Alus using the Fisher exact test to calculate the p-value and odds ratio. (K) Dot blot for dsRNA using total RNA from ID8 cells. (L) Scatter pot of signal intensity per cell detected by flow cytometry of ID8 cells immunostained for dsRNA (J2). IgG control used for threshold gating. (M) Fluorescence microscopy of ID8 cells immunostained for dsRNA.

Previous reports of de novo L1 integrants in ovarian HGSCs suggest L1 ORF1p and ORF2p must be active during HGSC development(26,27). Indeed, recent reports describe detection of secreted L1 ORF1p in plasma or ascites of HGSC patients(28,29). L1 ORF1p expressed in STICs is most likely derived from young L1 elements that are more likely to possess ORFs encoding reverse transcription machinery. Assessment of RNA-seq data reveals that the majority of LINEs expressed in STIC lesions derive from the L1 family and include full-length (~6kb) L1 elements (Figure 3C–D). L1s expressed in STIC lesions include young L1HS and L1PA2 elements (Figure 3E–F, Figure S4A–B). Accordingly, the majority of LINEs expressed in ID8 cells following p53 loss include L1 elements (Figure S4C).

Endogenous L1 ORF2p is nearly imperceptible in cancers with genetic features consistent with active L1 ORF2p(30). Recent work demonstrates that L1 ORF2p can generate cytosolic L1 reverse transcription products in part through cytosolic priming(16). To assess potential cytosolic accumulation of L1-derived agonists of DNA sensors following p53 loss, we returned to our cellular models. Total cytoplasmic nucleic acids from ID8 cells were subjected to RNaseI/III digestion followed by DNaseI treatment to isolate RNA specifically from cytoplasmic RNA:DNA hybrids (Figure S4D–F). RT-qPCR from the remaining RNA template reveals that p53-deficient ID8 cells harbor relatively elevated cytoplasmic levels of L1Md-A2-derived RNA:DNA hybrids that are no longer detected above wildtype levels following RNaseH treatment (Figure 3G). Accordingly, immunofluorescence confirms elevated cytosolic levels of RNA:DNA hybrids in p53-deficient cells (Figure S4G).

Both autonomously transcribed- and co-transcribed repetitive elements embedded within UTRs(13) and introns(31) can form immunostimulatory dsRNAs that bind cytosolic RNA sensors. RNA-seq reveals that STIC lesions express inverted repeat Alu (IR Alu) SINEs confirmed to form dsRNAs that bind the cytosolic RNA sensor MDA5(32) (Figure 3H–I) and ‘high force’ IR Alus predicted to form dsRNAs (bioRxiv 2021.11.04.467016) (Figure 3J). Likewise, p53 loss elevates expression of repetitive elements predicted to form dsRNAs in ID8 ovarian epithelial cells, OVE4, and OVE16 oviduct cells (Figure S4H–J). Examples of predicted dsRNA-forming sequences in murine cells include B2 SINEs that exhibit upregulation following p53 loss (Figure S4K). Therefore, we next assessed cytosolic dsRNA levels. Dot blots of total RNA (Figure 3K, Figure S4L) and flow cytometry using the dsRNA-specific J2 antibody (Figure 3L, Figure S4M) both reveal elevated dsRNA levels in p53 deficient cells. Immunofluorescence confirms elevated dsRNA signal in the cytosol of p53-deficient ID8 and OVE4 cells (Figure 3M; Figure S4N). Therefore, p53 loss elevates cytosolic dsRNA levels. Collectively, despite transcriptional changes of repetitive elements in both directions (Figure 1), p53 loss results in increased accumulation of potentially immunogenic nucleic acids that include RNA:DNA hybrids and dsRNAs.

Altogether, TP53 missense- and nonsense mutations are associated with elevated expression of L1 ORF1p, young L1s, and IR Alus that form dsRNA agonists of MDA5 specifically in STICs and not in adjacent normal epithelial tissue. Accordingly, p53 loss alone elevates cytosolic levels of L1-derived RNA:DNAs and cytosolic dsRNAs. Therefore, p53 loss elevates cytosolic levels of retrotransposon-derived nucleic acids with immunostimulatory potential to bind DNA- and RNA-sensors. Thus, downregulation of immunogenic dsRNAs or RNA:DNA hybrids does not explain reduced interferon induction in p53 deficient cells.

Constitutive dsRNA- or RNA:DNA exposure establishes viral mimicry tolerance

Accumulation of cytosolic RNA:DNAs and dsRNAs following p53 loss further underscores the paradoxical observation of evasion from pronounced viral mimicry induction. However, while acute IFN activation promotes antitumor responses, chronic IFN responses can conversely impart pro-tumorigenic effects through the promotion of immune evasion, a critical hallmark of cancer required for successful initiation of tumorigenesis(33,34). Therefore, we assessed whether constitutive dsRNA- or RNA:DNA-induced IFN responses following p53 loss might impact innate immune responses to confer pro-tumorigenic effects.

We sought to specifically assess contributions of dsRNAs towards transformation. Since dsRNA-sensing occurs in a sequence-independent manner, we used the synthetic dsRNA polyI:C to model the effects of constitutive dsRNA exposure. To model chronic agonism of dsRNA sensing, Trp53^+/+ ID8 cells were transfected with polyI:C, counted, then re-seeded at the original density for subsequent treatments. polyI:C concentrations were successively doubled with every treatment cycle (Figure 4A). After 6 treatment cycles, Trp53^+/+ ID8 cells were no longer sensitive to otherwise lethal doses of polyI:C as determined by live cell counts (Figure 4B). We then assessed molecular outputs of viral mimicry throughout treatment. Western blots of cytosolic fractions reveal that the lowest dose of polyI:C used at the first treatment cycle is sufficient to upregulate cytosolic RNA sensors MDA5 and RIG-I (Figure 4C). Accordingly, acute polyI:C treatment of unconditioned cells induces expression of Ifnβ and downstream ISGs that include Mx1 and Mx2, indicative of activation of the type I IFN response and viral mimicry induction (Figure 4D–F). Strikingly, with successive exposure to increasing concentrations of polyI:C, cells exhibit relatively diminished activation of the Type I IFN response. Indeed, cell populations that acquire complete dsRNA-insensitivity as determined by live cell counts exhibit no induction of the type I IFN response to markedly elevated dsRNA exposure (Figure 4D–F). This suggests that constitutive exposure to dsRNA selects for tolerance to elevated dsRNA levels characterized by diminished induction of the Type I IFN response. We refer to this dsRNA-insensitive condition as a state of ‘viral mimicry tolerance’.

Figure 4: Viral mimicry conditioning diminishes Type I IFN signalling.

(A) Schematic of cell conditioning approach (B) Viable cell counts of experimental groups throughout treatment course (C) Western blot of cytoplasmic fractions from ID8 cells following initial treatment cycle (D-F) Relative mRNA expression of indicated IFN response genes across treatment determined by RT-qPCR (G) Dot blot of total RNA from ID8 cells for dsRNA. Methylene blue stain of same membrane used as loading control. (H) Relative mRNA expression of indicated genes 24h post-siRNA transfection determined by RT-qPCR (I) Western blots of immunoprecipitations from nuclear extracts of ID8 cells. (J) Western blots of chromatin protein fractions from ID8 cells. (K-L) Heatmaps and average profiles of H3K27Ac CUT&RUN signal in Trp53^+/+ and Trp53^−/− ID8 cells at the promoter regions of all genes (n=24,974) (K) and at ISGs (n = 25) (L). Signal tracks were plotted with +/− 2Kb from the TSS of the genes (M) Genome tracks display H3K27Ac CUT&RUN signal in Trp53^+/+ and Trp53^−/− ID8 cells and IgG CUT&RUN signal in Trp53^+/+ ID8 cells at the Isg15 locus. (N) Western blots of whole cell extracts to assess sgRNA efficiency in Trp53^+/+ ID8 cells. (O) RT-qPCR for indicated targets 24h-post polyI:C- or mock treatment of Trp53^+/+ ID8 cells CRISPR-edited as indicated. * = p<0.05 by two-way ANOVA. (P) RT-qPCR for indicated targets 24h-post polyI:C- or mock treatment combined with dual p300/CBP inhibition using A485 in Trp53^+/+ ID8 cells. A486 used as negative control. * = p<0.05 by two-way ANOVA.

Since p53 loss elevates cytosolic RNA:DNAs, we conducted the same treatment course using the synthetic RNA:DNA pA:dT to assess the effect of chronic RNA:DNA exposure. Unlike dsRNA conditioning, chronic RNA:DNA conditioning did not affect ID8 cell viability (Figure S5A). However, increased treatment cycles established a similar dampening effect on the type I IFN response (Figure S5B). Therefore, constitutive agonism of dsRNA- or RNA:DNA sensing diminishes the Type I IFN response in association with increased tolerance of elevated cytosolic nucleic acid levels. We describe this selection process for diminished viral mimicry induction as ‘viral mimicry conditioning’.

We next assessed whether acute versus constitutive p53 loss induced similar changes in IFN signaling as observed during dsRNA- or RNA:DNA-conditioning. Acute p53 knockdown by siRNA elevates dsRNAs and Ifn⍺/β induction (Figure 4G–H). Importantly, knockout of the cytosolic DNA sensor cGAS abrogates ISG induction following acute p53 knockdown (Figure S5 C–E), suggesting that DNA sensing is required for viral mimicry induction upon acute p53 loss. In contrast, constitutive p53 deficiency increases cytosolic dsRNAs and RNA:DNAs but decreases expression of transcriptional programs associated with antiviral responses as revealed by Gene Set Enrichment Analysis (GSEA) of RNA-seq (Figure S2D–F). Specifically, Trp53^−/− ID8 cells exhibit reduced expression of ISGs that characterize the viral mimicry response (Figure S2E–F). This phenomenon is recapitulated in immortalized murine oviduct OVE4 cells. Acute p53 knockdown initially elevates Isg15 expression which subsequently decreases in expression with sustained p53 knockdown, and exhibits further downregulation to baseline levels upon removal of the siTrp53 treatment (Figure S5F–G). Viral mimicry tolerant (dsRNA-insensitive) ID8 wildtype cells exhibit sustained dsRNA insensitivity, as determined by cell survival, following re-exposure to polyI:C after prolonged cessation of polyI:C conditioning (Figure S5H–I). This suggests that some effects of viral mimicry conditioning are sustained following cessation of the conditioning process. Altogether, akin to p53 wildtype cells conditioned with chronic dsRNA or RNA:DNA exposure, constitutive p53 loss increases accumulation of cytosolic dsRNA and RNA:DNAs in concert with a paradoxical decrease in antiviral signaling to achieve a state of viral mimicry tolerance.

We sought to understand how p53-deficient cells reduce ISG and cytokine expression to tolerate elevated levels of cytosolic nucleic acids. We assessed protein levels of upstream effectors of the viral mimicry response that include cytosolic RNA- and DNA-sensors, along with downstream transcription factors. Western blots reveal reductions in cytosolic cGAS and modest reductions in RIG-I levels within p53-deficient cells (Figure S5J). Since cGAS and RIG-I are also ISGs(35), we assessed expression of downstream effectors of nucleic acid sensing responses such as STING and IRF7 that regulate ISG activation. Western blots reveal that p53-deficient cells do not exhibit reductions in STING or IRF7 protein expression relative to p53 wildtype cells (Figure S5K). Therefore, we focused upon downstream mechanisms of IRF-mediated IFN and ISG activation.

IRF transcription factors bind the histone acetyl transferases p300/CBP to facilitate transcription of IFNs and ISGs(36,37). Western blots of IRF7 immunoprecipitations performed using nuclear extract reveal that IRF7 binds p300, CBP, and IRF3 in both Trp53^+/+ and Trp53^−/− ID8 cells (Figure 4I). However, Western blots of chromatin fractions reveal reductions in chromatin-bound p300/CBP in p53-deficient cells (Figure 4J). Reductions in other histone modifying enzymes, such as SETD2 or SET, were not observed (Figure 4J). Importantly, dsRNA conditioning alone modestly reduces p300/CBP chromatin association (Figure S5L). Despite reduced p300/CBP chromatin occupancy, p53 deficient cells do not exhibit overt differences in H3K18Ac or H3K27Ac as determined by Western blots of chromatin fractions (Figure S5M). We conducted CUT&RUN to assess whether reduced p300/CBP-chromatin binding was associated with altered histone acetylation. CUT&RUN reveals that p53 loss does not reduce H3K27Ac genome-wide (Figure 4K). However, p53-deficient cells exhibit reduced H3K27Ac specifically at promoters of IFNs and ISGs involved in viral mimicry (Figure 4L–M).

To assess whether p300/CBP are required for viral mimicry responses, knockouts of either acetyltransferase were established in Trp53^+/+ ID8 cells (Figure 4N; Figure S5N). Knockout of either p300 or CBP diminished Ifn⍺ and Ifnβ induction in response to synthetic dsRNA treatment (Figure 4O). Treatments with the p300/CBP dual inhibitor A485 recapitulated this effect in the presence of either polyI:C or pA:dT (Figure 4P; Figure S5O). Therefore, p300/CBP are required for viral mimicry responses to dsRNA and RNA:DNA, and exhibit reduced chromatin occupancy in p53-deficient cells or dsRNA-conditioned cells. Reduced p300/CBP chromatin binding is associated with reduced H3K27Ac at ISGs that are silenced in viral mimicry-tolerant cell populations.

Collectively, we conclude that constitutive exposure to immunogenic dsRNAs or RNA:DNAs selects for diminished viral mimicry induction to facilitate tolerance of elevated cytosolic nucleic acid levels. We refer to this conditioning process as ‘viral mimicry conditioning’. Reduced ISG expression is associated with reduced CBP/p300-dependent H3K27Ac.

Viral mimicry tolerance impedes antitumor adaptive immunity

Tumor-intrinsic IFN responses promote secretion of cytokines that induce adaptive antitumor immune responses(38). We assessed whether diminished Type I IFN signaling established during viral mimicry conditioning impacts adaptive antitumor immune responses through reduced cytokine secretion. Akin to IFN and ISG mRNA expression, chronic dsRNA exposure in Trp53^+/+ cells reduces secretion of soluble IFN⍺/β and proinflammatory cytokines that include CCL4, CXCL10, and GMCSF involved in promoting paracrine antitumor responses (Figure 5A; Figure S6A–E). Likewise, Trp53^−/− cells exhibit reduced proinflammatory cytokine secretion relative to Trp53^+/+ cells following acute polyI:C treatment, consistent with viral mimicry tolerance (Figure 5B–C; Figure S6F). Therefore, viral mimicry-tolerant cells exhibit reduced proinflammatory cytokine secretion upon dsRNA exposure.

Figure 5: Acquisition of viral mimicry tolerance affects antitumor adaptive immune responses.

(A) Secreted cytokine concentration within culture media 24h-post treatment of Trp53^+/+ ID8 cells across treatment course. (B-C) Secreted cytokine concentration within culture media 24h-post treatment of ID8 cells. (D) Kaplan-Meyer survival curve of C57BL/6 female mice post-intraperitoneal injection of ID8 cells. p<0.0001 by Mantel-Cox test. (E) Cytokine concentration within ascites of ID8-injected C57/BL6 females at sacrifice. N=8 per group. * = p<0.05 by t-test. (F) Scatter plots display scRNA-seq dimensional reduction coefficients UMAP_2 versus UMAP_1 for ascites-derived CD45+ cells isolated at sacrifice from C57/BL6 females injected with Trp53^+/+ ID8 cells (left, n=12,615 cells) or Trp53^−/− ID8 cells (right, n=12,092 cells). Cell dots are color-coded according to corresponding cluster identity. Plots display cluster identity and corresponding cell type annotation per cluster. (G) Stacked bar plot displays proportions of single cells in each cluster from either experimental group. Counts of cells in each cluster from Trp53^−/− ID8-injected mice were compared with the count of all cells to calculate the odds ratio and p-value using the two-sided Fisher exact test. Y axis represents the proportion of cells in each dataset, while x axis represents the identified clusters numbered from 0–19. (H) Scatter plots of the scRNAseq dimensional reduction coefficients UMAP_2 versus UMAP_1 correspond to T cell clusters and display expression level per single cell for three T cell-associated markers which are, from left to right, Cd3e, Cd8a, and Cd4. Upper row represents cells from Trp53^+/+ ID8-injected mice while lower row represents cells from Trp53^−/− ID8-injected mice. (I) Pathway enrichment of markers that distinguish cluster 13 from other T-cell clusters.

We next assessed whether diminished proinflammatory cytokine secretion of p53-deficient cells in cell culture was sustained or altered in vivo. Intraperitoneal injection of p53-deficient ID8 cells into 6–8 week-old wildtype C57BL/6 female mice produces ascites onset sooner than isogenic p53 wildtype ID8 cells (Figure 5D), consistent with previous reports(39). Importantly, Trp53^+/+ and Trp53^−/− ID8 cells exhibit equivalent proliferation rates, suggesting that differences in ascites onset in vivo cannot be attributed to any potential difference in proliferative control (Figure S6G–H). Analysis of ascites collected at necropsy (Figure S6I–J) reveals that mice injected with p53-deficient cells exhibit reduced ascites concentrations of IFNβ and GMCSF (Figure 5E; Figure S6K). Therefore, reduced proinflammatory cytokine secretion associated with viral mimicry tolerance and p53 deficiency is recapitulated in vivo within the murine peritoneal cavity.

We assessed whether reduced cytokine secretion from Trp53^−/− ID8 cells impacted abundance or activation of cell populations that mediate adaptive anti-tumor immune responses. We conducted scRNA-seq of ascites-derived CD45+ cells isolated at sacrifice from C57BL/6 female mice injected with Trp53^+/+ ID8 cells or Trp53^−/− ID8 cells (Figure S6I–J). We identify 20 clusters of CD45+ cells across all biological samples (Figure 5F). Use of immune cell signatures to annotate cell types reveals that the majority of CD45+ cells within the peritoneal cavity are comprised of macrophages, neutrophils, B-cells, T-cells, and dendritic cells respectively (Figure S6L–N). Scatter plots of dimensional reduction coefficients UMAP_2 versus UMAP_1 reveal differences in CD45+ cell subsets based upon Trp53 genotype of the injected ID8 cells (Figure 5G). Notably, mice injected with Trp53^−/− ID8 cells exhibit pronounced reductions in all subsets of neutrophils and reductions in specific subsets of B-cells and macrophages corresponding to clusters 11 and 17 respectively (Figure 5F–G). Assessment of T cell subpopulations using known T-cell subtype markers reveals the presence of CD4 T-cells corresponding to cluster 5, and CD8 T-cells corresponding to clusters 4 and 13 (Figure 5H; Figure S6L–O). One of the two CD8+ T-cells clusters, cluster 13, exhibits reduced abundance in mice injected with Trp53^−/− ID8 cells (Figure 5G–H; Figure S6O). CD8 T-cell subpopulations in cluster 13 are distinguished from cluster 4 by enrichment of genes involved in signal transduction of T cell responses to non-self antigens (Figure 5I; Figure S6P).

Altogether, viral mimicry conditioning is associated with reduced secretion of proinflammatory cytokines. Accordingly, in contrast to acute p53 loss, constitutive p53 loss is characterized increased tolerance to dsRNA marked by reduced secretion of proinflammatory cytokines and concomitant reductions in adaptive antitumor immune responses associated with reduced survival.

Viral mimicry conditioning impacts therapeutic strategies

Induction of proinflammatory cytokines and activation of immune populations that respond to non-self antigens underlie the efficacy of epigenetic therapies, such as DNMT inhibitors (DNMTi), that function through viral mimicry induction(13). Genome-wide loss of DNA methylation following DNMTi treatment robustly elevates dsRNA levels to trigger viral mimicry that promotes adaptive antitumor immune responses in many preclinical cancer models(13). The reductions in cytokine secretion and immune cell populations involved in response to non-self antigens observed under conditions of viral mimicry tolerance begged the question of whether different viral mimicry sensitivities impact therapeutic response. Therefore, we assessed in vivo response to DNMTi in p53-deficient cells. To assess DNMTi response in vivo, Trp53^+/+ or Trp53^−/− ID8 cells were injected into the intraperitoneal cavity of 6–8 week-old wildtype C57BL/6 female mice that were subsequently treated with the DNMTi AZA or vehicle (Figure 6A, Figure S7A). Consistent with viral mimicry sensitivity, mice injected with Trp53^+/+ ID8 cells exhibit significantly increased survival following AZA treatment relative to vehicle-treated animals (Figure 6B). In contrast, mice injected with Trp53^−/− ID8 cells exhibit relative insensitivity to AZA treatment (Figure 6B) despite increased T-cell abundance within the peritoneal cavity (Figure S7B), consistent with increased viral mimicry tolerance.

Figure 6: Therapeutic impact of viral mimicry conditioning.

(A) Drug treatment schedule with intraperitoneal injections of 0.5 mg/kg of 5-Aza or 1x PBS as vehicle. (B) Kaplan-Meyer survival curve of C57BL/6 female mice post-intraperitoneal injection of ID8 cells stratified by treatment. (C) Plot of live cell count of Trp53^−/− ID8 cells across treatment time course. Cells were treated weekly with vehicle or 7.5μM 3TC. (D) Kaplan-Meyer survival curve of C57BL/6 female mice post-intraperitoneal injection of Trp53^−/− ID8 cells stratified by treatment. (E) Kaplan-Meyer survival curve of germline Trp53^−/− mice stratified by treatment group. Log-rank test p-value = < 0.0001. (F) Proposed model depicting acquisition of viral mimicry tolerance.

In agreement with scRNA-seq, characterization of ascites-derived immune cell populations at sacrifice reveals elevated CD4+ T-cell populations in mice injected with Trp53^−/− ID8 cells relative to mice injected with Trp53^+/+ ID8 cells (Figure S7C). However, CD4+ T-cells produce less IFNγ in Trp53^−/− ID8-injected mice (Figure S7D–E). Bulk characterization of CD8 T-cells reveals no appreciable differences in abundance based on ID8 cell genotype (Figure S7F–G), consistent with more subtle changes in specific CD8 T cell subsets detected by scRNA-seq. Instead, mice injected with Trp53^−/− ID8 cells exhibit modest reductions in IFNγ-positive CD8 T-cells at baseline (Figure S7H) and more pronounced reductions in IFNγ production following CD8 T-cell stimulation by phorbol myristate acetate (PMA) (Figure S7I). Altogether, p53-deficient ovarian epithelial cells exhibit increased resistance to DNMTi therapy in vivo that relies on cancer cell sensitivity to viral mimicry.

Recent work demonstrates that the pro-tumorigenic effects from chronic IFN(33) or chronic STING agonism(34) are reversable following genetic- or pharmacological disruption of IFN signal transduction. We sought to determine whether disruption of constitutive agonism of cytosolic nucleic acid sensing could disrupt viral mimicry conditioning that contributed to cancer cell immune evasion. Recent reports suggest that 3TC treatment of p53-deficient human cells can diminish proinflammatory signaling(21), with ongoing investigations suggesting that the anti-inflammatory effects of 3TC may be mediated through inhibition of L1 ORF2p(40) cytosolic priming(16) or the inflammasome(41). With recent publication of pilot phase II clinical trial results of 3TC treatment of TP53 mutant colorectal cancers(42), we assessed the impact of 3TC treatment on viral mimicry tolerant p53-deficient cells.

3TC treatment modestly decreases RNA:DNA levels (Figure S7J). However, Trp53^−/− ID8 cell growth (Figure 6C), along with expression levels of IFNs and ISGs (Figure S7K), are largely unaffected by weekly 3TC treatments at clinically-relevant doses relative to mock-treated groups over an 8 week treatment course. To assess in vivo efficacy, 6 week-old C57BL/6 females mice injected with Trp53^−/− ID8 cells were treated with 3TC administered at 2mg/ml ad libitum via drinking water to achieve clinically-relevant serum 3TC concentrations as previously described(17). 3TC-treated mice injected with Trp53^−/− ID8 cells did not exhibit significant survival differences relative to vehicle treated animals (Figure 6D).

Our mechanistic experiments suggest that prolonged conditioning by immunogenic cytosolic nucleic acids in p53-deficient cells gradually selects for diminished IFNs and ISGs involved in inducing antitumor immunity. Trp53^−/− ID8 cells that have undergone extensive passaging and selection in cell culture already exhibit tolerance to elevated cytosolic nucleic acid levels in association with diminished H3K27Ac at ISGs, consistent with viral mimicry tolerance. Therefore, we explored in vivo efficacy of 3TC treatment when administered at early stages of conditioning prior to transformation. Germline Trp53^−/− mice were treated with 2mg/ml 3TC ad libitum via drinking water beginning at P21 until animal protocol endpoints. 3TC-treated Trp53^−/− mice developed spontaneous lymphomas with no difference in penetrance compared to vehicle-treated mice. However, 3TC-treated Trp53^−/− mice exhibit increased tumor latency (Figure 6E). Thus, 3TC treatment confers efficacy when applied prior to spontaneous lymphomagenesis in vivo.

DISCUSSION

The majority of human cancers express repetitive elements(43). However, whether repetitive element RNAs represent a passive consequence or direct cause of tumorigenesis has remained unclear. In this study, we report that repetitive elements contribute towards the acquisition of cancer hallmarks. Specifically, we conclude that transcribed repetitive elements promote a type of tachyphylaxis that establishes cells capable of evading host immune defenses.

We report that premalignant STIC lesions exhibit pronounced transcriptional dysregulation of repetitive elements along with a paradoxical muted viral mimicry response. Cellular modeling to explore this paradox reveals that acute loss of p53 derepresses repetitive elements and suffices to induce viral mimicry (Figure 6F). However, prolonged agonism of RNA- and DNA-sensing diminishes viral mimicry induction to select for tolerance of elevated cytosolic levels of immunogenic nucleic acids. Thus, chronic agonism of cytosolic RNA- and DNA sensors promotes a form of tachyphylaxis that diminishes induction of ISGs and secretion of proinflammatory cytokines required to promote adaptive immune responses against cells with malignant potential. This selection process, that we refer to as viral mimicry conditioning, also diminishes response to viral mimicry-inducing DNMTi therapy.

While the findings of our study are broadly relevant to the majority of human cancers that inactivate the p53 pathway, we focus on early HGSC tumorigenesis due to the nearly 100% penetrance of TP53 mutations as a cancer-initiating event(3). In doing so, we obtain STIC-specific RNA-seq datasets to reveal the transcriptional changes that underlie HGSC premalignancy. Importantly, use of STICs from germline BRCA1/2 wildtype patients rules out confounding effects from BRCA1/2 mutations implicated in retrotransposon dysregulation in other settings. We conclude that both missense and nonsense TP53 mutations permit repetitive element dysregulation that precedes transformation. Associations between p53 function and repetitive element regulation have been noted in other model systems, and yet, mechanisms underlying this association remain unclear. We report that p53 occupies chromatin under non-DNA damaging conditions. We observe transcriptional dysregulation of repetitive elements in both directions following p53 loss in STICs, OVE cells, FTSECs, and ID8s. While this study focuses on upregulated repetitive elements capable of stimulating cytosolic nucleic acid sensors, observations of downregulated repetitive elements following p53 loss represents an interesting future direction to pursue. Differences in the numbers of repetitive elements dysregulated across cells from different species may be explained by differences in the number and conservation of repetitive elements across species(44). Curiously, repetitive elements that become dysregulated upon loss of p53 exhibit diminished H3K9me3, but lack p53 occupancy in wildtype cells. Therefore, we suggest that the sensitivity of repetitive element silencing to p53 status may be explained by indirect mechanisms involving p53 regulation of H3K9me3 that require further clarification. Whether H3K9me3-mediated silencing of repetitive elements exhibits p53 sensitivity in other cancers requires direct investigation.

Identifying a mechanism of cancer immune evasion through chronic exposure to cytosolic nucleic acids expands upon an emerging body of literature that reveals the agonists of protumorigenic inflammatory responses(33,34). Previous studies suggest that cytosolic DNA may arise, in part, from DNA damage associated with the activity of young L1 retrotransposons in p53-deficient cancer cells(45,46). However, cytosolic nucleic acids in cancer cells can originate from multiple sources beyond expressed repetitive elements that include ssDNA or dsDNA from ruptured micronuclei or RNA:DNA hybrids from R-loops. Importantly, sequence agnostic recognition of cytosolic nucleic acids by RNA- and DNA-sensors suggests that the potential agonists of viral mimicry conditioning may not be limited to retrotransposons, but direct investigations are required. The translational discoveries from this emerging body of literature challenge the use proinflammatory therapies in cancers characterized by prolonged periods of chronic IFN induction. Indeed, we report reduced DNMTi efficacy in cells that exhibit increased dsRNA tolerance. Instead, cancers that undergo selection in the presence of chronic IFN may be better suited towards strategies that attenuate chronic IFN responses, as reported using genetic approaches(33), STING inhibitors(34), or JAK inhibitors(47). This approach has been demonstrated in non-cancer models as illustrated by the use of 3TC to inhibit accumulation of immunostimulatory L1-derived reverse transcription intermediates in TREX1-deficient Aicardi-Goutières syndrome (AGS) models(48). Likewise, the use of a reverse transcriptase inhibitor cocktail inhibits L1-mediated cGAS stimulation following dendritic cell activation and subsequent H3K9me3 depletion at LINEs (bioRxiv 2021.09.10.457789). Accordingly, we explore the use of 3TC as a potential strategy to attenuate viral mimicry conditioning. Importantly, the anti-inflammatory effects of 3TC reported by multiple groups require detailed investigations to clarify whether therapeutic efficacy is mediated through L1 ORF2p inhibition(40) or inhibition of the inflammasome(41).

One of the key limitations of our study includes the lack of therapeutic assessments in a spontaneous model of HGSC derived solely from Trp53 mutations in the tube. Spontaneous murine HGSC models that most closely resemble human HGSC pathogenesis require additional mutations beyond Trp53(49,50) that would confound the ability to determine the consequence of p53 loss on viral mimicry initiation and viral mimicry tolerance in isolation. The development of spontaneous murine HGSC models with only Trp53 mutations that recapitulate key features of human HGSC pathogenesis would be of great interest for future studies. In this study, use of ID8 cells to study mechanisms of tumor initiation represents a study limitation since Trp53 mutations are not required for ID8 tumorigenicity(39).

A second major consideration is that STICs used in this study present with concurrent HGSC tumors. Incidental or ‘pure’ STIC cases devoid of concurrent HGSC tumors are predominantly collected from prophylactic surgeries of germline BRCA1/2 mutant carriers. This study specifically explores the contribution of p53 loss or mutation in isolation. Incidental STICs with BRCA1/2 mutations would confound this analysis because BRCA1/2 mutations are independently associated with transcriptional dysregulation of repetitive elements(51). Therefore, STIC lesions devoid of BRCA1/2 mutations were selected to determine the effect of TP53 mutations in isolation unconfounded by germline BRCA1/2 mutations. The majority of such STICs are discovered in patients with concurrent HGSCs. However, transcriptomes from treatment-naïve patients with incidental STICs from non-BRCA1/2 mutant carriers devoid of concurrent HGSCs would provide the clearest insight into transcriptional changes present during the earliest stages of HGSC initiation. Multi-institutional efforts to collect and share such transcriptomes will be valuable for future studies of this topic.

Altogether, uncovering contributions of repetitive elements towards cancer-initiation can reveal strategies to guide therapy selection and opportunities to expand upon new therapeutic strategies suited towards cancers that develop in the presence of chronic IFN induction.

METHODS

Cell Culture

Adherent ID8 cells were cultured in DMEM (Gibco) supplemented with 12U/ml penicillin, 100μg/ml streptomycin, 292 μg/ml L-glutamine, 4% fetal bovine serum (ThermoFisher 10378016), and (Gibco) ITS (5 mg/mL insulin, 5 mg/mL transferrin, and 5 ng/mL sodium selenite). CRISPR modified ID8 cells used in this study were a kind gift from Dr. Iain McNeish (University of Glasgow) and included Trp53^+/+ and Trp53^−/− (F3 clone) cells(39). 28–2 ID8 cells were a kind gift from Dr. Jim Petrik (University of Guelph)(52). Trp53^+/+ and Trp53^−/− OVE4 cells cultured in this study were a kind gift from Dr. Trevor Shepherd and were cultured as previously described(53). Briefly, adherent OVE4 cells were cultured in Alpha Modification of Eagle’s Medium 1X (AMEM) (Corning^™) supplemented with 5% FBS (Life Technologies), MEM non-essential amino acids (Fisher Scientific), 0.01 mg/mL insulin-transferrin-selenium solution (Gibco), 0.01 μM β-estradiol (Sigma-Aldrich), and 2 ng/mL human EGF (Shenandoah Biotechnology). Cells were cultured and maintained at 37°C and 5% CO₂ with routine testing to confirm lack of mycoplasma infection.

Animal Models

Mouse strains used for this study include C57BL/6 (University Health Network Animal Resources Centre) and B6.129S2-Trp53^tm1Tyj/J (The Jackson Laboratory). Intraperitoneal injections of 5×10⁶ ID8 cells in 0.2ml PBS were performed on 6–8 week C57BL/6 females. Necropsies were performed upon animal protocol endpoint as determined by veterinary staff of the University Health Network Animal Resources Centre. Genotyping of B6.129S2-Trp53^tm1Tyj/J mice was performed as per distributor’s protocol by The Centre for Applied Genomics at SickKids. Necropsies were performed upon animal protocol endpoint as determined by veterinary staff of the University Health Network Animal Resources Centre.

For in vivo 3TC treatments, 3TC (Cayman Chemical) was maintained at 2mg/ml in drinking water and changed twice per week at day 3 and day 7 by the University Health Network Animal Resources Centre. Animal necropsies were typically performed by the University Health Network Animal Resources Centre. Animal tissues archived by FFPE were sectioned and H&E-stained by the UHN Pathology Research Program core facility. Animal tissue pathology analysis was conducted by Dr. Ming Tsao. Animals were housed and handled as approved by the Canadian Council on Animal Care with extensive technical support from the University Health Network Animal Resources Centre for both animal husbandry and experiments.

siRNA Knockdown

For siRNA transfections, cells were seeded for ≤50% confluence next day in antibiotic-free media. Next day, siRNA-lipid complex was generated using RNAiMAX (Invitrogen) as per manufacturer’s protocol in OptiMEM media (Gibco), applied to cells for a final siRNA concentration of 10nM, then incubated at 37 °C for at least 24 hours prior to harvesting for RNA or protein. siRNAs used in this study include p53 siRNA (m) (Santa Cruz sc-29436) and control siRNA-A (Santa Cruz sc-37007).

CRISPR Knockout

CRISPR gene editing was performed using the Alt-R CRISPR-Cas9 System (Integrated DNA Technologies) as per manufacturer’s protocol. Alt-R^® CRISPR-Cas9 crRNA was mixed with Alt-R^® CRISPR-Cas9 tracrRNA and Alt-R^® HiFi S.p. Cas9 Nuclease V3 to assemble the ribonucleoprotein (RNP) complex that was subsequently electroporated into target cells using pulse code EH115 and Lonza P3 primary cell 4D-Nucleofactor^™ X Kit S (Basel, Switzerland). Alt-R^® CRISPR-Cas9 Negative Control crRNA #2 was used to generate non-targeted isogenic controls. Specific gene knockouts were generated using guide RNAs listed in Table S2 that were selected using the IDT Alt-R^® CRISPR-Cas9 guide RNA design and selection tool. Polyclonal populations were used for experiments following CRISPR targeting of ID8 cells with the following guide RNAs: sgCGAS, sgEP300, sgCREBBP.

Patient Cohort Selection

All work using clinical samples was conducted according to protocols subjected to institutional review and approval (UHN REB 19–5917). Clinical records of corresponding FFPE-archived STIC cases from the UHN Laboratory Medicine Program (LMP) archives were assessed to identify incidental STIC cases that lack of pathogenic germline BRCA1/2 mutations. Presence of pathogenic TP53 and BRCA1/2 mutations were determined from the concurrent tumour for each case by the UHN Advanced Molecular Diagnostics Laboratory (AMDL). Briefly, DNA was extracted from FFPE using a Maxwell 16 FFPE plus LEV DNA purification kit (Promega, Madison, WA). To detect BRCA1, BRCA2 or TP53 mutations, extracted genomic DNA was sheared using focused-ultrasonification (Covaris LE220, Woburn, MA). 250ng of sheared DNA was subjected to target enrichment using the SureSelectXT Target Enrichment System (Santa Clara, CA), followed by sequencing on the NextSeq500 (Illumina) platform. Alignment and variant calling was performed by the Burns-Wheeler Alignment, GATK and VarScan2. Variants not meeting laboratory defined quality metrics (read depth <100, population frequency >1% in the gnomAD database, or variant allele fraction <5%) were removed from further analysis. Variants of uncertain significance (VUS) were not considered pathogenic. Cohort sample characteristics appear in Table S1.

Patient tissue sectioning

FFPE blocks of suitable incidental STIC cases were retrieved from the UHN Laboratory Medicine Program (LMP) archives, then sectioned by the UHN Princess Margaret Cancer Biobank (PMCB) and stained by the UHN Pathology Research Program (PRP) core facility as follows: 17 serial sections were prepared from each sample. Top and bottom sections were cut at 4μm and stained with haematoxylin and eosin (H&E) by UHN PRP, then reviewed by our pathology team to confirm the presence of STIC lesion and lack of concurrent HGSC invasion into the lesion when relevant throughout each series. Five 4μm sections were designated for downstream immunohistochemistry, performed by UHN PRP. Ten 10μm sections were mounted on RNA membrane slides (PEN membrane 2μm coated glass slides, DNAse and RNAse free, Leica #11505189) for downstream laser capture microdissection (LCM) by UHN PMCB.

Patient tissue immunohistochemistry

4μm sections designated for immunohistochemistry were stained for Ki67 (Dako #M7240, clone MIB1), p53 (Leica, NCL-p53-DO7 clone D07), or L1 ORF1p (22) (Millipore Sigma, MABC1152 clone 4H1). Deparaffinization was conducted at room temperature followed by 3% H₂O₂ treatment at room temperature for 15 minutes to block endogenous peroxidase activity. Antigen retrieval was conducted with Tris-EDTA pH 9.0 for Ki67 staining, or Citrate pH 6.0 for either p53- or L1 ORF1p staining. Blocking was conducted at room temperature using 2.5% normal horse serum. Ki67 primary antibody staining was conducted at 1/300 dilution overnight at room temperature. p53 primary antibody staining was conducted at 1/1000 dilution for 1 hour at room temperature. L1 ORF1p primary antibody staining was conducted at 1/300 dilution overnight at room temperature. MACH 4 (Intermedico Cat# BC-MU534G) detection kit was used followed by DAB (3,3’-diaminobenzidine) chromogen (DAKO Cat# K3468) and Mayer’s Hematoxylin counter stain for 15 seconds at room temperature.

Patient tissue microdissection and RNA isolation

H&E-stained sections were scanned by UHN LMP Pathology Client Services for electronic demarcation of STIC and adjacent normal regions of interest by our pathology team. These were used to guide regions to be captured by LCM. Following demarcation, RNA membrane slide mounted sections were deparaffinized using CitriSolv^™ (Decon Labs) and stained with a light hematoxylin counter stain by UHN PRP. Stained slides were then microdissected by UHN PMCB. Pre- and post-LCM images were captured to confirm specific isolation of lesion and adjacent normal tissue. LCM resulted in capture of STIC and adjacent normal tissue in separate tubes. Buffer was added to each tube and tissue samples were subsequently stored at −80°C. Dual RNA/DNA extraction from FFPE samples was performed by Tissue Portal (TP) within the Diagnostic Development Program at OICR. A semi-automated extraction method using the ThermoFisher KingFisher Flex system was used. The extraction protocol was specifically optimized for this work using adjacent tissue from the FFPE sections. Isolated RNA was quantified using a Qubit Fluorometer.

Patient tissue RNA-seq

25ng of isolated RNA from FFPE tissue was subjected to rRNA depletion and library preparation using the Illumina^® Stranded Total RNA Prep, Ligation with Ribo-Zero Plus kit (Illumina, cat# 20040529) with 13 cycles of amplification used. Prepared libraries were assessed for quality and fragment size distribution using Agilent High Sensitivity DNA kit (Agilent, cat# o 5067–4627) on the Bioanalyzer 2100 system. Where relevant, dimers were removed using AMPure XP reagents (Beckman, cat# A63880). Libraries were then sequenced on an Illumina MiSeq to confirm sufficient unique alignment rates to hg38 as a validation of rRNA-depletion and sufficient library complexity. Libraries were subsequently sequenced on an Illumina NovaSeq6000 to obtain ≥80M reads per sample. RNA-seq data shown in this report corresponds to the following STIC cases within our cohort: STIC0001, STIC0002, STIC0004, STIC0007, STIC0008, STIC0009, STIC0010, and STIC0011.

polyI:C and pA:dT conditioning

Cells were seeded for 50% confluency next day in a 6 well dish. Next day, cells were transfected with vehicle or either polyI:C (Invivogen) or pA:dT (Invivogen) using the LyoVec transfection reagent (Invivogen) as per manufacturer’s protocol. 24 hours later, media was collected and 2× 1ml fractions were stored at −80°C for downstream cytokine quantification. Cell culture media was replaced with fresh culture media. 24 hours after media collection, viable cells were counted, then re-seeded at the original seeing density. Remaining cells were frozen as cell pellets and stored at −80°C for downstream RNA extraction. With every treatment cycle, polyI:C or pA:dT concentration was doubled.

3TC treatment in cell culture

Cells were seeded for at 3.125×10⁴ cells per well in a 6 well dish in media containing vehicle or 7.5uM 3TC. On day 4, media was replaced with fresh media containing vehicle or drug at same concentration. On day 7, viable cells were counted, then re-seeded at the original seeing density in media containing vehicle or 7.5uM 3TC to begin a new weekly treatment cycle. Remaining cells were frozen as cell pellets and stored at −80°C for downstream RNA extraction.

Immunofluorescence

ID8 cells were seeded in 8-well coverglass-bottom chambers (Eppendorf) to achieve 50% confluency at fixation. To fix, cells were washed 3x in 1x PBS, incubated in 4% paraformaldehyde while rocking for 10 minutes, then washed 3x in 1x PBS. To permeabilize, cells were incubated in 1x PBS-T (0.3% Triton X-1000 in 1xPBS) for 10 minutes while rocking. Cells were then incubated in blocking buffer (1% BSA in 1x TBS-T) rocking for 1 hour at room temperature to block. Blocking buffer was replaced with primary antibody diluted in blocking buffer and incubated overnight rocking at 4°C. Cells were then washed 3x in 1x PBS-T, incubated in secondary antibody (isotype-matched IgG-Alexa Fluor 647) diluted in 1:800 in 1x PBS-T rocking at room temperature for 1 hour while protected from light. Cells were then washed 3x in 1x PBS-T, then incubated in ActinGreen^™ 488 diluted in PBS-T as per manufacturer’s protocol rocking for 30 minutes while protected from light for experiments in which Actin was stained for. Cells were then washed 3x in 1x PBS-T. Coverslip was detached from the 8-well chamber and mounted onto a glass slide using SlowFade^® Gold antifade reagent with DAPI (Life Technologies). Samples were cured overnight on a flat dry surface at room temperature while protected from light. Next day, coverslip sealant was applied to seal edges of coverslip to glass slide. Slides were stored protected from light until image capture by fluorescence- or confocal microscope. Summary of all antibodies used in this study appears in Table S3.

Cell culture confluence assay

ID8 cells were seeded overnight at 300 cells per well in a 96 well plate. Next day, drug or vehicle was added and cells were scanned in an Incucyte Kinetic Imaging System (Essen BioScience) every 3 hours to capture culture confluence. Upon completion of time course, treatment confluence values were normalized against the vehicle control confluence values and presented as a percentage of control sample. At the endpoint, all cell samples were incubated with PrestoBlue^™ Cell Viability Reagent (Thermo Fisher Scientific, A13261) according to the manufacturer’s protocol to determine relative cell viability.

Cell fractionation

Cellular fractionation was adapted from Mendez and Stillman (Méndez and Stillman, 2000). Cells were resuspended at 4×10⁷ cells/ml in Buffer A (10 mM Tris [pH 8.0], 10 mM KCl, 1.5 mM MgCl2, 0.34 M sucrose, 10% glycerol, 1 mM DTT) supplemented with protease/phoshatase inhibitors and incubated on ice for 5 minutes. An equal volume of 0.3mg/ml digitonin in Buffer A was added followed by a 10 minute incubation on ice. 5–10% total reaction volume was then collected as a whole-cell fraction. Following centrifugation at 1300 ×g for 5 minutes at 4°C, supernatant containing the cytoplasmic fraction was transferred to a new tube. Pelleted nuclei were washed in Buffer A, then incubated 30 minutes on ice in a volume of Buffer B (3 mM EDTA, 0.2 mM EGTA, 1 mM DTT, fresh protease/phosphatase inhibitors) ~2x that of the pellet to promote lysis. Following centrifugation at 1700 ×g for 5 minutes at 4°C, supernatant containing the nucleoplasmic fraction was transferred to a new tube. The chromatin pellet was resuspended in DNaseI buffer (20mM Tris [pH 7.5], 10 mM MgCl₂) at ~2x pellet volume, supplemented with 200U of DNaseI. Following a 1 hour incubation on ice, all fractions were snap frozen and stored at −80°C for subsequent protein quantification. Western blots were conducted as per standard protocols. Histone loading controls were detected by staining polyacrylamide gels with GelCode^™ Blue Stain Reagent.

Dot Blot

Total protein lysate was normalized across all samples prior to RNA extraction. RNA samples extracted using the TRIzol manufacturer’s protocol were pipetted onto a positively charged nylon membrane (Amerhsam Hybond-N⁺) membrane in a maximum volume of 3uL/sample. Samples were air dried for 10–15 minutes at room temperature, then UV crosslinked. Membrane was blocked for 1 hour in blocking buffer (5% milk diluted in 0.01% PBS-T). Blocking buffer was discarded, and membrane was incubated in primary antibody diluted in blocking buffer rocking overnight at 4°C. Membrane was washed 3x in PBS-T rocking for 10 minutes at room temperature per wash. Membrane was then incubated in HRP-conjugated secondary antibody diluted in blocking buffer rocking 1 hour at room temperature. Membrane was washed 3x in PBS-T rocking for 10 minutes at room temperature per wash. Washed membrane was subjected to chemiluminescent development. To detect total nucleic acid loading, membrane was then incubated 0.5% methylene blue in 30% EtOH rocking 1 hour at room temperature or overnight at 4°C. After incubation, membrane was washed with distilled water, and stained membrane was imaged.

Immunoprecipitation

Immunoprecipitation protocol was adapted from Ishak et. al.(54). Briefly, thawed extract was cleared of cellular debris by centrifugation at 20,800 g for 30 min at 4 °C. 0.5–1.5mg of cleared extract was used per immunoprecipitation (IP). Cleared extracts between experimental groups were normalized between experimental groups using lysis buffer. 0.5–1.5mg of extract was used per IP. Normalized extract was diluted in IP dilution buffer (20mM Tris pH 7.5, 0.2 mM EDTA, and 0.1% NP-40) supplemented with protease/phosphatase inhibitors to achieve final NaCl concentration ≤ 200mM. 5–10% of IP volume was designated for input assessments. Protein complexes were immunoprecipitated with 5ug of antibody pre-bound to 30uL of washed protein A/G Dynabeads (Invitrogen) per IP. IPs were conducted ≥ 1 hour at 4°C, washed twice with IP wash buffer, (20mM Tris pH 7.5, 200 mM NaCl, 1.5 mM MgCl2, 0.2 mM EDTA, 25 mM DTT and 0.1% NP-40), eluted in Laemmli sample buffer by boiling, then isolated from beads and transferred to new tubes for western blot analysis.

RIPA whole cell extract preparation

Generation of whole-cell lysates was adapted from Ishak et. al. (54). Briefly, PBS-washed cells were scraped into 1x PBS and pelleted by centrifugation at 900 g for 5 min at 4°C. Cells were resuspended in ice-cold RIPA buffer (50 mM Tris-HCl [pH 8.0], 150 mM NaCl, 1% Triton X-100, 0.5% sodium deoxycholate, and 0.1% SDS) supplemented with protease/phosphatase inhibitors at ~2x pellet volume and incubated for 20 minutes on ice. Lysates were snap frozen and stored at −80°C for subsequent processing. Thawed lysates were cleared of cellular debris by centrifugation at 20,800 g for 20 min at 4 °C. Supernatants were transferred to new tubes for protein quantification and boiled 5 minutes in Laemmli sample for downstream Western blotting.

Restriction digests

All restriction digests were performed as per manufacturer’s protocol (New England Biolabs).

Isolation of cytosolic RNA:DNA hybrids.

Cytosolic cellular fractions were conducted as described in ‘Cell fractionation’ method. 5mg of total cytosolic extract per genotype was prepared in equal volume normalized with Buffer A (Cell fractionation). Cytosolic protein fractions were digested with proteinase K in 0.5ml reactions with the following reaction conditions: 1mg of total cytosolic protein in 350μL of Buffer A, 150mM NaCl, 0.5% SDS, 100 μg/mL proteinase K. Reactions were incubated ≥1 hour at 37–40°C to digest protein. Digestion was the subjected to standard phenol chloroform extraction followed by ethanol precipitation to isolate total cytoplasmic nucleic acids from digested protein.

RNA:DNA hybrids were isolated from total cytosolic nucleic acids by restriction digest using a modified protocol adapted from Tharp et. al. (55). RNaseIII and RNaseI were selected for the first restriction digest to remove dsRNA and ssRNA respectively. Enzymes were then added to the digestion reactions. Double digestion reactions were prepared as follows: 1 μg cytosolic nucleic acids, 1x NEB ShortCutBuffer, 20mM MnCl₂, RNaseIII (NEB M0245S) at 1U enzyme / 1 μg nucleic acids, RNaseI (NEB M0243L) at 1U enzyme / 0.5 μg nucleic acids and sterile H₂O up to 10uL total volume per reaction. 2% of total reaction volume was removed and stored as input prior to addition of enzymes. Reactions were incubated for 30 minutes at 37°C, then supplemented with 10mM MgCl₂ for RNaseI, and incubated for 30 more minutes. A total of 6 RNaseIII and RNaseI double digestion reactions were conducted per genotype. Upon completion of incubation, reactions from the same genotype were combined at 2:1 for 3 reaction tubes total per genotype, and subjected to phenol-chloroform extraction and ethanol precipitation to isolate undigested nucleic acids. Precipitated nucleic acids were resuspended in a final volume of 20uL in sterile H₂O.

The resuspended pools of nucleic acids were then divided in two equal volumes for an RNaseH digestion and a ‘mock’ digestion. RNaseH digestion reactions were prepared as follows: 50% total volume of nucleic acids isolated from RNaseI/III digest, 1x NEB RNaseH buffer, RNaseH (NEB M0297L) at 5U enzyme / reaction and sterile H₂O up to 25uL total volume per reaction. RNaseH was omitted for ‘mock’ reaction. Reactions were incubated for 30 minutes at 37°C. Immediately upon completion of incubation, reactions from all groups were supplemented for DNaseI digestions as follows: 25uL of previous RNaseH (or mock) digestion, 1X NEB DNaseI buffer, sterile H₂O up to 50uL final volume. Reactions were incubated for 10 minutes at 37°C. Nucleic acids from digests and input samples were then isolated using the QIAgen RNeasy Micro Kit (QIAgen 74004) as per manufacturer’s protocol. Isolated RNA was subjected to reverse transcription and RT-PCR as described in methods and quantified as % input. Primer sequences in 5’−3’ orientation used for RT-qPCR to detect L1Md-A2 ORF1(56) are as follows: L1 ORF1_F (AGATCTGGAACCATAGATG), L1 ORF1_R: TTCTCATTGTGTCCTGGATT.

RT-qPCR

Total RNA isolation and DNaseI treatment was performed using RNeasy Mini Kit (Qiagen) as per manufacturer’s protocol. Reverse transcription of 1ug of RNA / sample was performed using SuperScript Vilo IV (Thermo Fisher Scientific) as per manufacturer’s protocol. Resulting cDNA was diluted 1/10 in nuclease-free sterile and used in RT-qPCR reactions SYBR MasterMix. All RT-qPCR reactions were performed in a CFX Connect Real-Time PCR Detection System (Bio-Rad) and consisted of 1X SsoAdvanced Universal SYBR Green Supermix (Bio-Rad), 0.3 mM of forward and reverse primers, 1uL of diluted cDNA and sterile deionized water to complete the final volume of 10 μL. Gene expression values were calculated by the ΔCq method, using Rplp0 as the housekeeping gene, and resulting experimental target values were normalized to the global mean of the control group. Normalized fold change was plotted using GraphPad Prism Version 6.0 or 8.0 software. Summary of all primers used in this study appears in Table S4.

CUT&RUN

CUT&RUN protocol and all buffer recipes used were adapted from Skene et. al.(57) and performed in 0.2ml tubes. Digitonin concentrations were selected based upon trypan blue permeabilization assessment conducted one day prior to experiment as previously described (57). Briefly, 200,000 cells were washed and immobilized on 10uL of activated Concanavalin A-coated magnetic beads per condition. Captured cells were resuspended in 0.2ml of antibody buffer containing primary antibody diluted at 1:100. Tubes were rotated 2 hours at 4°C to promote antibody binding to target protein. Cells were then washed with 0.2ml of digitonin wash buffer and resuspended in 0.2ml of digitonin wash buffer containing pA-MNase (expressed and purified in house) diluted to a final concentration of ~0.8ng/uL. Tubes were rotated 1 hour at 4°C to promote protein A binding to primary antibody. Buffer was discarded and replaced with 150uL of digitonin wash buffer. Tubes were cooled to 0°C for 5 minutes, then supplemented with CaCl₂ to promote MNase digestion. After 30 minutes, 150uL of 2X STOP buffer containing chelating agents and heterologous yeast spike-in DNA (kind gift Dr. Steven Henikoff, Fred Hutchinson Cancer Research Center) was added to stop the reaction. Cells were incubated at 37°C for 10 minutes to promote release of soluble chromatin fragments, then centrifuged at 16,000×g for 5 minutes at 4°C to pellet cells. Supernatant containing soluble chromatin fragments was transferred to new tubes for DNA extraction. Phenol chloroform extraction was conducted on fragments bound by transcription factors to capture small fragments that may not be captured by column-based extraction methods. Column-based extraction (QIAgen minElute) was conducted for histone CUT&RUN experiments since minimum fragment sizes from histone CUT&RUN experiments typically exceed minimum ranges for column-based DNA extraction. DNA libraries were prepared by PMGC using either NEBNext Ultra II DNA library kit (New England Biolabs) or ThruPLEX DNA-seq kit (Takara Bio USA) as per manufacturer’s instructions. Unique molecular indexes were included with p53 CUT&RUN experiments. Libraries were sequenced to obtain ≥40M paired-end reads / sample on an Illumina NovaSeq6000.

Bulk RNA-seq

For freshly cultured cells, total RNA isolation and DNaseI treatment was performed using RNeasy Mini Kit (Qiagen) as per manufacturer’s protocol. Library preparation was conducted using the TruSeq^® Stranded Total RNA Library Prep Gold (96 Samples) kit (Illumina, 20020599). Sequencing was conducted on an Illumina HiSeq2500.

Isolation of ascites-derived cells

Freshly euthanized animals were subjected to necropsy. With peritoneal cavity intact, ascites fluid was extracted with a 20 gauge needle fixed to a 10ml syringe. Ascites fluid was immediately transferred to 10ml EDTA-coated tubes that were centrifuged at 600×g for 10 minutes at 4°C in a swing bucket rotor centrifuge. Cleared ascites fluid was discarded, and cell pellet was resuspended in 5ml 1x ACK buffer (150mM NH₄Cl, 10mM KHCO₃, 0.1 mM Na₂EDTA) to lyse red blood cells. Following 1 minute incubation, cells were centrifuged 350×g for 5 minutes at 4°C. Supernatant containing lysed red blood cells was discarded, and 1x ACK treatment followed by centrifugation was repeated once more. Remaining cells were washed with 1x PBS to remove ACK buffer.

Staining of Ascites-derived T-cells for flow cytometry

Cells were transferred from EDTA-coated tubes to 96 well plates, seeded at a density of 1×10⁶ cells / well. Cells were then washed on 0.2ml FACS buffer (2% FBS or 0.5% BSA in 1xPBS). To identify viable cells, cells were stained with LIVE/DEAD^® Fixable Violet Dead Cell Dye (ThermoFisher Scientific, catalog No. L-34963). Extracellular surface markers were subsequently stained at 4°C for 30 minutes. Antibody dilutions were typically 1:100 in FACS buffer. Cells were centrifuged for 5 minutes at 300–400xg, and supernatant containing antibodies was discarded. Stained cells were fixed by incubation in 0.1ml Fixation Buffer (ThermoFisher Scientific, catalog No. 00–5523-00) for 30 minutes at 4°C protected from light.

For intracellular marker staining, cells were stimulated with Cell Stimulation plus Protein secretion inhibitor Cocktail (ThermoFisher Scientific, catalog No. 00–4975-03) for 5 hours prior to staining with cell viability dye and cell surface markers as described above. Cells were washed 2x in 0.2ml of 1x Permeabilizing Buffer (ThermoFisher Scientific, catalog No. 00–5523-00). Cells were centrifuged for 5 minutes at 300–400xg, and supernatant was discarded. Cells were then stained 30–60 minutes at room temperature with intracellular antibodies. Cells were washed 1x in 0.2ml of 1x Permeabilizing Buffer, washed 2x in 0.2ml of FACS buffer, then resuspended in 0.2ml FACS buffer and stored at 4°C until flow cytometry analysis. Analysis of results was conducted using FlowJo software for downstream histogram generation.

CD45+ cell isolation for scRNA-seq

To isolate CD45+ cells from ascites for scRNA-seq, ascites cell isolation was performed as described above, and staining of extracellular markers was conducted as described above directly in polypropylene tubes. Live cells were immediately subjected to FACS by the SickKids-UHN Flow Cytometry Facility. At least 4000 sorted cells / group were then immediately subjected to 10x Genomics library preparation by the Princess Margaret Cancer Genomics Centre for sequencing on an Illumina HiSeq2500.

Cell Proliferation Assay

A total of 25,000 Trp53^+/+ and Trp53^−/− ID8 cells were stained using the CellTrace^™ Violet (CTV) Cell Proliferation Kit (Cedarlane, catalog No. 423114) and seeded into 6- and 12-well plates. Cells were cultured for 0, 1, 2, or 4 days, then harvested and washed in 2 mL FACS buffer (2% FBS or 0.5% BSA in 1x PBS). To identify viable cells, cells were stained with SYTOX^™ Green Dead Cell Stain (ThermoFisher Scientific, catalog No. S34860) for flow cytometry. Cells were centrifuged at 200 × g for 5 minutes. CTV dilutions of 1:2000 in FACS buffer were used for a 10-minute incubation, after which the supernatant containing CTV was discarded. The live and dead stain dilutions were also prepared at 1:2000 in FACS buffer. Live cells were collected and analyzed using a BD FACSymphony^™ A3 Cell Analyzer at the SickKids-UHN Flow Cytometry Facility, with at least 10,000 live cells recorded per group.

Cytokine Quantification

1ml of conditioned media was speed-vacuumed to 0.1ml (10x reduction) and subjected to cytokine quantification using the LEGENDplex^™ Mouse Anti-Virus Response Panel (13-plex) with V-bottom Plate and LEGENDplex^™ Mouse Proinflammatory Chemokine Panel (13-plex) with V-bottom Plate as per manufacturer’s protocol.

In vivo DNMTi treatment

Mouse DNMTi treatment experiments were performed in accordance with the approved protocol by the Institutional Care and Use Committee (IACUC) at The George Washington University (Protocol A406). C57B/6 mice were obtained from the Charles River Laboratories (Wilmington, Massachusetts, USA). For in vivo tumor studies, ID8 Trp53^−/− or Trp53^+/+ isogeneic CRISPR modified ID8 murine ovarian epithelial cells, a kind gift from Dr. Iain McNeish, were cultured and expanded in DMEM with GlutaMAX-I (Gibco, 10569–010), 4% FBS, 1X Insulin-Transferrin-Selenium (Gibco, 41400–045), and 1X Pen/Strep (Gibco, 15070–063).

Intraperitoneal injections were performed using 5.0×10⁶ ID8 cells suspended in 500 uL 1x phosphate buffered saline (PBS) (Corning 21–040-CV). Drug or vehicle treatment began 1 week after tumor inoculation and ended when tumor burden was deemed excessive as per animal protocol. All mice were injected with drug or vehicle 5 days per week, every other week until experiment completion. DNMTi treatment consisted of 0.5mg/kg of 5-azacytidine (Sigma-Aldrich, A2385) dissolved in 100 μl of 1X PBS solution per injection. Vehicle treatment group consisted of intraperitoneal injections of 100 μl of 1X PBS.

Tumor burden was assessed via measurement of body weight, ascites volume drained upon 20%-30% body weight increase, or abdominal circumference measurement. Statistical outliers were removed using Pierce’s criterion. 10 mice were designated per experimental group. However, final group numbers were affected by animals that succumbed to ID8 treatment or had not yet developed ascites by designated ascites collection time points. Furthermore, differences in biological replicates obtained for flow cytometry experiments were affected by limited ascites-derived cell yields from some animals.

Flow cytometry of cells from DNMTi-treated mice

Ascites was collected from 5–10 mice per group and then centrifuged to pellet cells. Supernatant was removed and then ACK Buffer was added to the cell pellet, inverting tubes for 2 minutes to lyse red blood cells. To neutralize the ACK buffer and to wash the cells, PBS was added and then samples were centrifuged. The ACK lysis and wash was repeated once more. Ascites cells for the myeloid panel were resuspended in RPMI with 10% FBS and 1% Pen/Strep. Ascites cells for the T cell panel were then resuspended in RPMI with 10% FBS, 1% Pen/Strep and Cell Stimulation Cocktail plus Protein Transport Inhibitors at the manufacturer recommended concentration (eBioScience, 00497503) and then incubated for 4 hours at 37°C and 5% CO₂.

For the T cell panel, 1×10⁶ cells per sample were washed and then stained with: Live/Dead Fixable Aqua (Thermo Fisher Scientific, L34965), anti-CD45-PerCP/Cy5.5 (BioLegend, 103132), anti-CD3-AlexaFluor700 (BioLegend, 152316), anti-CD4-BV785 (BioLegend, 100453), anti-CD8a-APC/Fire750 (BioLegend, 100766). Cells were fixed after viability and surface staining using the fixation buffer from Foxp3/Transcription Factor Staining Buffer Set (eBioscience, 00–5523-00). On the day of acquisition, cells were permeabilized using the 1X Permeabilization Buffer from the Foxp3/Transcription Factor Staining Buffer Set (eBioscience, 00–5523-00). Intracellular/nuclear staining was performed with anti-IFN-γ-FITC (BioLegend, 505805).

All flow cytometry data acquisition was performed using a 3-laser, 12-color BD Celesta analyzer. All flow cytometry data were analyzed using the FlowJo Software 10.6 and statistical outliers were removed using Pierce’s criterion.

Statistical analysis of flow cytometry from DNMTi treated mice

Mouse experiments were done in duplicate, and Kaplan-Meier survival curves were conducted using GraphPad Prism 7. Unpaired t-tests were performed on flow cytometry data using GraphPad Prism 7.

Analysis of CUT&RUN data without UMIs

Fastq reads were trimmed using fastp program v. 0.19.5(58) to remove adapters and remove bases with low quality score. The fastp command was run with the options -l 30, q 30 and paired-end mode. Trimmed reads were aligned to the mouse genome mm10 using bowtie2 tool v. 2.3.5(59) using the same alignment setting used in (Skene et al., 2018)(57), which used the following option setting:

-q -I 10 -X 700 --local --very-sensitive-local --no-mixed --no-discordant --no-unal --phred33

Duplicated reads were kept in the downstream analysis, while the unaligned reads and discordantly aligned reads were discarded. The trimmed reads were also aligned to the S. cerevisiae yeast genome (sacCer3) to evaluate the yeast spike-in reads for the spike-in normalization using the same bowtie2 parameters as described previously (Skene et al., 2018)(57), which used the following option setting:

-q -I 10 -X 700 --local --very-sensitive-local --no-mixed --no-discordant --no-overlap --no-dovetail --no-unal --phred33

Alignment files (in sam format) were converted to bam format and sorted by genomic coordinates using samtools v. 1.9(60). Bam files of the replicates of each condition were merged using samtools to create genome coverage tracks. Whole genome coverage tracks in bigwig file format were created from the merged bam files of the replicates using the bamCoverage command from the deepTools v.3.5.0(61). The bamCoverage command was used with the options --scaleFactor f, --extendReads, --binSize 5, and --skipNonCoveredRegions, where f is a normalization scale factor. f was calculated as the ratio of 20000 to the number of reads aligned to the S. cerevisiae yeast genome (sacCer3). This normalization is used to normalize the samples for the spike-in DNA used in the CUT&RUN samples as described previously (Skene et al., 2018)(57). Heatmaps of the genomic regions were plotted from the bigwig files using computeMatrix and plotHeatmap commands in the deepTools package.

Analysis of CUT&RUN data with UMIs

CUT&RUN samples that include universal molecular identifiers (UMI) with their read sequences are preprocessed by UMI-Tools v. 1.0.0(62). UMIs (6 bp) were extracted from the reads and added to the read names using the extract command from the UMI-Tools package as follows:

umi_tools extract --bc-pattern=NNNNNN --bc-pattern2=NNNNNN -I Input_Read1_Fastq -S Output_Read1_Fastq --read2-in=Input_Read2_Fastq --read2-out=Output_Read2_Fastq

We then removed the STEM sequences which are 8 – 11 nt sequences on each end of the inserts from reads using custom python script. Reads were trimmed using fastp program to remove any adapters or base pairs with low quality scores. The fastp command was used with the options -l 25 and -q 30. Samples were then aligned to the mouse genome mm10 and S. cerevisiae yeast genome (sacCer3) using bowtie2 tool with the same command settings in (Skene et al., 2018)(57). The bowtie command was used to align the samples to the mouse genome mm10 and S. cerevisiae yeast genome (sacCer3) with the same sets of options described in the previous section. Aligned reads are sorted based on the genomic coordinates using samtools. Sorted reads were then grouped based on their UMI and genomic coordinates with allowing 1bp editing distance to connect UMIs using group command of the UMI-Tools. The group command was used with the options --edit-distance-threshold=1, --paired, and --unpaired-reads=discard. This command generates an output bam files that we do not use in the downstream analysis. It also generates a flat file (tsv file) that describes the read groups. Sorted reads were deduplicated based on the UMI and the mapping coordinates with allowing 1bp editing distance to connect UMIs using the dedup commands from the UMI-tools as follows:

umi_tools dedup --paired --unpaired-reads=discard --edit-distance-threshold=1 --stdin=Input_Sorted_Bam_file --log=logfile > Output_Bam_file

The deduplicated bam files are used for the downstream analysis such as peak calling and genomic tracks generation. Peaks are called from the pooled replicates deduplicated bam files using MACS2 v. 2.2.5(63) with the normalization of the samples using the yeast spike-in reads quantification. p53 peaks were called using the narrow peak mode. The macs2 callpeak command was used with the options:

--ratio r -g mm --keep-dup all -f BAMPE -q 0.05

where r is the normalization factor that is calculated as the ratio of the number of the p53 CUT&RUN sample that are aligned to the sacCer3 genome to the number of reads of the IgG sample that are aligned to the sacCer3 genome.

To conduct differential analysis for the binding of the p53 peaks between the Trp53^+/+ and Trp53^−/− conditions, we merged the p53 peaks from the Trp53^+/+ and Trp53^−/− conditions using merge command in the bedtools suite v.2.27.1(64). We converted the bed file of the merged peaks into GTF format. We quantified the reads for the merged peaks from the bam files of the replicates using the featureCount command(65) from the subread tools (v.1.6.2). We performed the differential analysis between the Trp53^+/+ and Trp53^−/− condition using edgeR library(66). Deferentially binding peaks were determined based on the cutoff FDR<0.05 and |log2FC|>1. Wildtype-specific peaks, which are the peaks that are downregulated in Trp53^−/− versus Trp53^+/+, have been intersected with the whole genome repeat element loci to find the overlap of these peaks with repeat elements. A bed file of whole genome repeat element loci which were marked by RepeatMasker as repeat element loci was downloaded from the UCSC genome browser table. The bed file was used to find the intersection of these loci with the wildtype-specific peaks. Heatmaps of the genome tracks of each repeat family that intersect with wildtype-specific p53 peaks were plotted for the Trp53^+/+ and Trp53^−/− conditions and other histone marks profiles for the both Trp53^+/+ and Trp53^−/− conditions were plotted using computeMatrix and plotHeatmap commands in the deepTools package.

Motif Analysis

De novo motif analysis was performed on all wildtype-specific p53 CUT&RUN peaks using the findMotifsGenome command in HOMER v. 4.8(67). All the rest of the p53 CUT&RUN peaks that are not wildtype-specific peaks were used as background in the motif search. The findMotifsGenome command was used with the options -len 8,10,12,20 and -size given. We also performed de novo motif analysis on wildtype-specific p53 CUT&RUN peaks that overlap with ERV1 repeats. These peaks usually overlap with the 5’ end of these ERV1 repeats. We used same regions of other ERV1 repeats that have no p53 CUT&RUN signal as background in this motif search. The findMotifsGenome command was used with the options -len 8,10,12, 14 and -size given.

GREAT Analysis

Analysis of the functional cis-regulatory regions was performed on the wildtype-specific peaks using GREAT v.4.0.4(68). The bed file of wildtype-specific peaks (n=2,127) was used as test regions and the whole genome was set to be the background regions in web server tool (http://great.stanford.edu/public/html/index.php). The settings option was selected to be “Single nearest gene” within 5.0Kb.

Bulk RNA-seq data analysis of mouse cells

Raw fastq files were trimmed using trim_galore tool v. (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/)) to remove adapters and low-quality bases. The trim_galore tool was used with the options –fastqc, --paired, --length 30, and -q 30. Trimmed reads were aligned to the mouse genome (mm10) using STAR tool v. 2.5.2(69). We allowed multiple-aligned reads that are aligned up to 100 loci to be used in the downstream analysis. This number of loci is recommended by the TEtranscript tool (70). The STAR alignment was done with the options --sjdbOverhang 99, --outFilterMultimapNmax 100, --outFilterMismatchNmax 3, --winAnchorMultimapNmax 200, --alignEndsType EndToEnd, --alignIntronMax 1, --alignMatesGapMax 350, and --seedSearchStartLmax 30.

Alignment files were sorted using samtools. Duplicate reads were marked using picard tools v.1.9.1(http://broadinstitute.github.io/picard/) and removed using samtools. Genome tracks bigwig files were created or the forward and the reverse strands from the alignment files using the bamCoverage command from the deepTools package. The genome track files were created for the merged replicates of each condition. The bamCoverage command was used with the following oprtions: --scaleFactor factor, --binSize 10, and --skipNonCoveredRegions, where factor is the normalization factor which was calculated as one million per the count of reads in the library (1000,000/library size). The command also included the option --filterRNAstrand forward for calculating the genome track for the forward strand and --filterRNAstrand reverse for calculating the genome track of the reverse strand.

Read counts were quantified for repeat elements and genes using TElocal tool which is a locus-wise version of TEtranscripts2(70). TElocal command was used with the options --stranded reverse, --sortedByPos and --mode multi. A GTF file for mm10 ensemble genes and a TElocal index for repeat elements were used with the TElocal command.

RNA-seq differential analysis at repeat elements was performed to assess the up/downregulated repetitive elements. Differential analysis was performed on the read counts using the edgeR library. The set of upregulated repeats was intersected with the mm10 set of regions that have high dsRNA force (bioRxiv 2021.11.04.467016) to find the repeats that have high dsRNA force and are upregulated in Lesion vs Normal tissue samples.

Differential analysis was performed on the subfamily level of the repeat elements using the TEtranscript2 tool. The TEtranscript command performs the read count quantification for the repeat subfamilies and genes and perform the differential analysis using DESeq2 R package. The TEtranscript command is used with the options --stranded reverse, --sortedByPos and --mode multi. Differentially regulated subfamilies were assessed as they have FDR<0.05 and abs(logFC)>1.

Read counts were quantified for genes using the featureCount command from the subread tools v. 1.6.2. The featureCount command was used with the options -M, -p, -B, -s 2, and –largestOverlap. An mm10 ensemble GTF file for the gene transcripts was used with the featureCount command. Differential analysis was performed on the gene counts to evaluate the differentially expressed genes. Genes that have FDR<0.05 and abs(logFC)> 1 are considered differentially regulated genes.

Bulk RNA-seq data analysis of human samples

Raw fastq files were trimmed using trim_galore tool v. (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/)) to remove adapters and low quality bases. The trim_galore command was used with the options: --paired, --length 40, and -q 30. Trimmed reads were aligned to the human genome (hg38) using STAR tool v. 2.5.2(69). Multiple-aligned reads that are aligned up to 100 loci were allowed to be used in the downstream analysis. This number of loci is recommended by the TEtranscript tool. The STAR alignment command was used with the options: --outFilterMultimapNmax 100, --outFilterMismatchNmax 3, --winAnchorMultimapNmax 200,--alignEndsType EndToEnd, --alignIntronMax 1 and --alignMatesGapMax 350. Alignment files were sorted using samtools. Duplicate reads were marked using picard tools v.1.9.1 and removed using samtools.

Genome tracks bigwig files were created for the forward and the reverse strands from the alignment files using the bamCoverage command from the deepTools package. The genome track files were created for the merged samples of each tissue type (Normal and Lesion). The bamCoverage command was used with the following oprtions: --scaleFactor factor, --binSize 10, and --skipNonCoveredRegions, where factor is the normalization factor which was calculated as one million per the count of reads in the library (1000,000/library size). The command also included the option --filterRNAstrand forward for calculating the genome track for the forward strand and --filterRNAstrand reverse for calculating the genome track of the reverse strand.

Read counts were quantified for repetitive element loci and genes using TElocal tool which is a locus-wise version of TEtranscripts2. TElocal command was used with the options --stranded reverse, --sortedByPos and --mode multi. A GTF file for hg38 Genecode v. 28 genes and a TElocal index for repetitive elements were used with the TElocal command.

RNA-seq differential analysis at repeat elements was performed to assess the up/downregulated repeat elements. Differential analysis was performed on the read counts using the edgeR library. Repetitive elements that have FDR<0.05 and abs(logFC)>1 are considered differentially regulated.

Differential analysis was performed on the subfamily level of the repeat elements using the TEtranscript2 tool. The TEtranscript command performs the read count quantification for the repeat subfamilies and genes and perform the differential analysis using DESeq2 R package. The TEtranscript command is used with the options --stranded reverse, --sortedByPos and --mode multi. Repeat subfamilies that have FDR<0.05 and abs(logFC)>1 are considered as differentially regulated.

Read counts were quantified for genes using the featureCount command from the subread tools v. 1.6.2. The featureCount command was used with the options -M, -p, -B, -s 2, and -O. An hg38 Gencode V.28 GTF file for the gene transcripts was used with the featureCount command. Differential analysis was performed on the gene counts to evaluate the differentially expressed genes. Genes that have FDR<0.05 and abs(logFC)> 1 are considered differentially regulated genes.

The set of upregulated Alus was intersected with the sets of while genome annotated and experimentally-validated MDA5-binding IR Alus(32) to find the IR Alus that are upregulated in Lesion vs Normal tissue samples. The set of upregulated repeats was intersected with the set of hg38 regions that have high dsRNA force (bioRxiv 2021.11.04.467016) to find the repeats that have high dsRNA force and are upregulated in Lesion vs Normal tissue samples.

Bulk RNA-seq analysis of public dataset

Human RNA-seq samples from GSE182487(14) (71) were downloaded from the GEO repository. Reads were trimmed to remove adapters and low-quality bases using the trim_galore tool. The trim_galore command was used with the options --paired, --length 40, and -q 30. Trimmed reads were aligned to the human genome hg38 using STAR aligner. Reads that are aligned to multiple loci are allowed to the downstream analysis if they are aligned to a maximum of 100 loci. The STAR command was used with the following options: --outFilterMultimapNmax 100, --outFilterMismatchNmax 5, --winAnchorMultimapNmax 200,--alignEndsType EndToEnd, --alignIntronMax 1 and --alignMatesGapMax 350. Aligned reads were filtered to remove PCR duplicated reads using picard and samtools. Read count quantification for repetitive elements was performed using the TElocal tool. TElocal command was used with ther options --stranded reverse, --sortedByPos and --mode multi. A GTF file for hg38 Genecode v. 28 genes and a TElocal index for repetitive elements were used with the TElocal command. RNA-seq differential analysis at repeat elements was performed to assess the up/downregulated repeat elements. Differential analysis was performed on the read counts using the edgeR library. Repetitive elements that have FDR<0.05 and abs(logFC)>1 are considered differentially regulated.

Raw data files of Mouse RNA-seq samples from GSE227681(53) were downloaded from the GEO repository. Reads were trimmed using trim_galore tool to remove adapters and low-quality bases. The trim_galore tool was used with the options –fastqc, --paired, --length 30, and -q 30. Trimmed reads were aligned to the mouse genome (mm10) using STAR tool. Multiple-aligned reads that are aligned up to 100 loci were allowed to be used in the downstream analysis. The STAR alignment was done with the options --sjdbOverhang 99, --outFilterMultimapNmax 100, --outFilterMismatchNmax 3, --winAnchorMultimapNmax 200, --alignEndsType EndToEnd, --alignIntronMax 1, --alignMatesGapMax 350, and --seedSearchStartLmax 30. RNA-seq differential analysis at repeat elements was performed to assess the up/downregulated repeat elements. Differential analysis was performed on the read counts using the edgeR library. Repetitive elements that have FDR<0.05 and abs(logFC)>1 are considered differentially regulated.

Donut Plots Analysis

The counts of upregulated repeat elements for each repeat class were calculated, and their proportions were displayed in a donut plot. To determine statistical significance, the counts of upregulated elements in each class were compared to the total counts of each class in the whole genome, calculating the p-value and odds ratio using Fisher’s exact test. Additionally, genomic annotation of the upregulated repeat elements was conducted, and the proportions of these elements in the SINE, LINE, and LTR classes across different genomic regions were also illustrated in donut plots. The counts of upregulated elements in a specific class within each genomic region were compared to the counts of elements of that class in the same region across the whole genome, using Fisher’s exact test to calculate the p-value and odds ratio.

Repeat Family-Based Quantification

Reads were mapped to the hg38 genome for human samples and to the mm38 genome for mouse samples using STAR v2.7.7a(69) taking into account known splice sites from Gencode(72) annotation. STAR mapping parameters were set to default. Gene counts were computed based on Gencode(72) annotation using featureCounts (v1.4.6-p5)(65). Repeats counts per element subfamily were primarily quantified against Dfam using featureCounts for uniquely mapped reads. Multimapping reads which were not assigned to any gene (Gencode annotation) or repeat (Dfam annotation) were mapped to consensus sequences of repeat elements (Dfam)(73). Repeat counts of a given subfamily is the sum of the counts obtained in each of these two stages.

scRNA-seq Data Analysis

10X Genomics scRNA-seq datasets were processed by cellranger pipeline(74) using the command count to generate single cell feature counts for each library. This command creates matrix of feature counts for the cells. We processed the feature matrices of the datasets with Seurat package(75–78) to perform QC and normalization of the multiple datasets. We used R script tools to merge the datasets and perform differential analysis(79). We performed QC normalization. We then merged and integrated the normalized datasets to correct the batch effects. The integrated datasets are saved in an R object. We then performed dimensional reduction and clustering of the integrated datasets. We did global cell clustering that includes global cell clustering in all the integrated datasets(79). In this step, UMAP/tSNE plots are plotted for global cell clusters, dataset, and dataset type (which include control and treatment types).

We used singleR library(80) to predict the types of the clustered cells. The normalized gene expression matrix of the cells was used to compare with ImmGenData dataset which is a mouse immunogenic dataset that includes genes sets for mouse immunogenic cell types. We then counted the cells associated with each cell type in each cluster. Statistical analysis was performed to evaluate the significance of the association of cell type with each cluster using Fisher exact test. A bar plot showing the log₂(Odds ratio) and the significance of the p-value for the clusters with their associated cell type was plotted using ggplot2 library. We also plotted a stacked bar plot showing the proportion of the cells that were associated with each cluster from the control and treatment datasets. In order to show the proportions of cells that were identified by each cell type from the control and treatment datasets, we plotted a stacked bar plot to show these proportions from the two datasets for each cell type.

scRNA-seq Differential Analysis

To detect markers that are more expressed in a cell type or cluster versus other cell types or clusters, we performed differential analysis using SUERAT package. In cell type differential analysis, we compared the expression of cells from the Trp53^+/+ and Trp53^−/− datasets collectively that were identified as a particular cell type versus the rest of cells. Significance of the upregulation of markers in cells of a particular cell type versus the rest of cells was determined by log₂FC>=3, adj_pvalue<0.00001, and the minimum percentage of cells in which the marker is expressed in either of the two groups of cells that were compared is 0.25. We detected 115 markers that were differentially expressed in cells of each cell type versus the rest of cells. We calculated and represented the expression of the markers in each cell type in a form of a dot plot by calculating the average expression of each marker in each the cells of each cell type and by calculating the percentage of cells (pct) in which the marker was expressed in cells of each cell type.

In another differential analysis, we compared the Trp53^−/− cells in each cluster of the four clusters that were identified as T cells with every other cluster of this cell type to detect markers that could represent subtypes of T cells. Significance of upregulation of markers in cells of each cluster versus cells of each other cluster of the T cells clusters was determined by log2FC>1, adj_pvalue<0.01, and the difference in the percentage of cells (pct1) in which the marker is expressed in each cluster and the percentage of cells in which the marker is expressed in the cells of each other cluster (pct2) should be greater than 0.3. The range of pct1 and pct2 is usually between 0 and 1. In this analysis, we detected about 40 markers that were upregulated in a cluster versus at least one other cluster. The average expression of the marker in the cells of each of the two clusters that were compared as well as the percentages of the cells of the two clusters in which the marker was expressed was calculated and depicted in a dot plot. The dot plot displays two dots for each cluster comparison where one dot represents the expression of the marker in the cluster where it is upregulated, and the other dot represents the expression of the marker in the other cluster where it is downregulated.

Statement of Significance.

Our landmark discovery of viral mimicry characterized repetitive elements as immunogenic stimuli that cull cancer cells. If expressed repetitive elements cull cancer cells, why does every human cancer express repetitive elements? Our report offers an exciting advancement towards understanding this paradox and how to exploit this mechanism for cancer interception.

Data Availability Statement

Sequencing datasets generated in animal cells for this study are publicly available in the NCBI Gene Expression Omnibus (GEO) at SuperSeries GSE216947. Sequencing datasets generated using human patient samples for this study are publicly available in the European Genome-phenome Archive (EGA) at EGAS50000000200. Sequencing datasets analyzed in this study that were originally generated in previous studies are publicly available in the NCBI Gene Expression Omnibus (GEO) at GSE227681 and GSE182487. Analysis code associated with this manuscript is accessible via Zenodo at https://zenodo.org/records/10139208.