Recurrent ZC3H18 mutations stabilize oncogenic endogenous retroviral RNA Authors: Tanzina Tanu, Anna M. Cox, Jennifer A. Karlow, Priyanka Sharma, Kaiwen Sun, Xueyang He, Swathy Babu, Madeline K. Roelofs, Ethan Jeon, Jiajia Chen, Constance Wu, Jared Brown, Nelson B. Barrientos, Andrew S. McCallion, Kevin M. Brown, Stephen J. Chanock, David Liu, Tongwu Zhang, Kathleen H. Burns, Paul L. Boutz, Megan L. Insco Published: 2026 ABSTRACT SUMMARYEndogenous retroviral (ERV) RNA is highly expressed in cancer, suggesting a selective advantage. We identify recurrent truncating mutations in ZC3H18 (Z18), a nuclear RNA surveillance factor lost in 30% of all cancers. We show that Z18 truncating mutations are oncogenic and that Z18 plays an evolutionarily conserved role in the nuclear surveillance of oncogenic ERV RNA. In zebrafish, Z18 truncation accelerates melanoma onset and selectively increases ERV RNA. Human cancer cell lines and patient tumors with Z18 mutations also upregulate ERV RNA. In engineered human melanoma cells, Z18 truncation enhances ERV RNA accumulation more than heterozygous Z18 loss, consistent with dominant-negative activity. Truncated Z18 directly stabilizes and relocalizes ERV RNA to the cytoplasm. ERV RNA expression accelerates melanoma and is required for Z18 truncation-mediated melanoma onset in zebrafish. Growth of human melanoma cells similarly depends on Z18-regulated ERV RNA. Together, these findings support Z18-regulated ERV RNA as a driver of oncogenesis. INTRODUCTION As healthy cells require appropriate levels of high-quality RNA, dysregulation of RNA homeostasis contributes to diverse human diseases, including cancer. RNA homeostasis requires balanced production and degradation. While the role of RNA production through transcription factors, chromatin regulation, and enhancers in cancer is well studied,1 how perturbations in RNA degradation pathways contribute to cancer is less clear. Mechanisms that detect and degrade aberrant or unstable nuclear RNAs have only recently been described,2–5 and we previously discovered that the deficient nuclear RNA surveillance of aberrant prematurely terminated RNAs (ptRNAs) is oncogenic.6 Nuclear RNA surveillance components are mutated in up to 21% of melanomas, and recurrent mutations occur across cancers in two components, including ZC3H186 (Z18). This pattern of mutations suggests that deficient nuclear RNA surveillance is a widespread contributor to human tumorigenesis. Different protein complexes target unstable or aberrant nuclear RNA for degradation. The polyA exosome targeting (PAXT) complex targets aberrant polyadenylated transcripts,4,7 including oncogenic ptRNAs.6 The nuclear exosome targeting (NEXT) complex3 identifies a wide range of noncoding RNAs for degradation, including promoter upstream transcripts (PROMPTs), enhancer RNAs (eRNAs),8 long interspersed element-1 (LINE-1) retrotransposons,9 and long terminal repeat (LTR) retrotransposons, including endogenous retroviral (ERV) RNAs.10 Retroviral-derived ERVs and LTRs comprise approximately 8% of the human genome11 and have accumulated mutations that render them noninfectious and immobile.12 ERV sequences can retain enhancer and polymerase II (Pol II) promoter activity, as well as splice sites and polyadenylation signals,13,14 enabling the generation of noncoding RNAs or novel open reading frames either independently or when incorporated into host transcripts.15 Transposable elements are upregulated in multiple cancer types, with the ERV class of retroelements being most affected.16,17 ERV upregulation in cancer correlates with increased immune infiltration and immunotherapy efficacy in multiple cancers.17–19 However, the causes of ERV upregulation in cancer are unknown, and it remains unclear whether elevated levels of ERV transcripts or proteins are pathogenic or merely an epiphenomenon of broader dysregulation in malignant tissue. In this study, we show that recurrent truncating mutations in Z18 produce a truncated protein that retains its ability to bind RNA and lacks the domain required to recruit RNA degradation machinery. This truncated isoform accelerated melanoma onset in a zebrafish model and specifically promoted the accumulation of ERV RNAs, without affecting the levels of other aberrant or unstable nuclear RNAs. We find that in humans, the role of Z18 in ERV degradation is conserved, that Z18 truncating mutations function in a dominant-negative manner, and that the truncated isoform binds and stabilizes ERV RNA in the cytoplasm. Importantly, expression of zebrafish or human ERV RNA was sufficient to accelerate melanoma in zebrafish, and ERV RNA was required for both mutant Z18-driven melanoma in zebrafish and the growth of human melanoma cells. By studying Z18 mutations in melanoma, we reveal a functional role for ERV RNA in oncogenesis. This study supports that accumulated aberrant RNA contributes directly to cancer phenotypes. RESULTS Recurrent Z18 truncating mutations are oncogenic We previously showed that cyclin-dependent kinase 13 (CDK13) activates the PAXT complex, which targets aberrant ptRNAs for degradation. When CDK13 is mutated, ptRNAs are stabilized, leading to more aggressive melanoma.6 We decided to investigate Z18, a PAXT component previously identified as a low-frequency driver in melanoma (OncodriveFM).6 In The Cancer Genome Atlas (TCGA) melanoma20 patient samples, PolyPhen-221 predicted enrichment of deleterious Z18 mutations (p = 0.0047). In a larger melanoma cohort (n = 1347), 35% of patients with copy-number data had bi-allelic or monoallelic loss (291/823). Z18 deletions were also enriched in the all-cancer dataset (Q = 1.75E−54, 30.0% frequency), which includes melanoma (Q = .0001, 36.6% frequency).22 Of the 64 melanoma patients with Z18 mutations, 50 (78%) had deleterious mutations (7 truncating and 43 missense) (PolyPhen-2) (Data S1, “Z18 Mut & Melanoma Patient Refs”). Other TCGA cancers also showed significant enrichment of deleterious mutations by PolyPhen-2, including uterine corpus endometrial carcinoma (UCEC) (p = 0.0041), cervical squamous cell carcinoma (p = 0.028), colon adenocarcinoma (COAD) (p = 0.013), and lung adenocarcinoma (p = 0.016) (Data S1, “PolyPhen-2 Results”). These data suggest selection of deleterious Z18 mutations and copy-number loss in melanoma and other cancers. We next analyzed Z18 mutations in a larger pan-cancer cohort (Data S1, “Pan-cancer Patient Sample Refs”). We identified 506 deleterious Z18 mutations from 23,411 patients and found that truncating mutations (Z18trunc) were statistically enriched in the region preceding the RNA surveillance complex-binding domain (amino acid 679–901) compared to the rest of the protein (odds ratio = 5.89, p = 1.66e–17, Fisher’s exact test) (Figure 1A; Data S1, “Z18trunc Fisher’s exact test” and “Pan-cancer Z18 Mutations”). Truncating mutations clustered at R680 (49 patients) and just downstream (37 patients), suggesting that loss of the distal C-terminal domain confers a selective advantage. Z18trunc mutations were most frequently found in endometrial carcinoma (20/67), stomach adenocarcinoma (18/29), and colorectal adenocarcinoma (14/35). Z18 mutations were enriched in microsatellite instability (MSI)-high tumors, although most occurred in microsatellite-stable cancers (Table S1). Because loss-of-function mutations are expected to be uniformly distributed across a gene, the enrichment of Z18 truncations suggests an alternative function, such as dominant-negative activity. The Z18 C-terminal domain is essential for recruiting RNA surveillance machinery.23 Thus, we hypothesized that these mutations create an oncogenic isoform that cannot degrade target RNAs, thereby promoting cancer. As human and zebrafish Z18 are highly similar (67.62% nucleotide sequence and 62.9% protein sequence using the tool BLAST24), we tested the function of the most frequent patient truncating mutation, Z18 R680Gfs*5 (hereafter Z18trunc), in zebrafish melanoma (Figure S1A), using the MAZERATI rapid genetic modeling system in BRAFV600E; p53−/−; mitfa−/− zebrafish (triple-mutant),25–27 which allows melanocyte-specific genetics in first-generation animals. Z18trunc expression produced more black patches (early melanoma) at 9 weeks (Figures 1B and S1B) and expedited melanoma onset compared to control eGFP expression (Figure 1C, replicate in S1C and S1D). Human Z18trunc expression accelerated melanoma onset in zebrafish. Z18trunc promotes ERV RNA accumulation We asked whether Z18trunc stabilizes RNA surveillance substrates. Because Z18 binds to the PAXT complex,4 which clears polyadenylated ptRNAs,6 we quantified ptRNAs from pA-selected bulk RNA sequencing (RNA-seq) from zebrafish tumors with melanocyte-specific expression of eGFP (n = 3) or Z18trunc (n = 3) (Figure S2A). A curated set of CDK13-dependent ptRNAs was not significantly upregulated (Figure S2B). As Z18 is also reported to associate with the NEXT nuclear RNA surveillance complex,3 we assayed NEXT substrates. Transposable element (TE) expression from pA-selected RNA-seq was quantified using a tool called SQuIRE.28 Average TE expression was calculated for eGFP (n = 3) and Z18trunc melanomas (n = 3), revealing a significant increase in LTRs/ERVs in Z18trunc-expressing tumors (p = 4.5e–12, Wilcoxon test) (Figure 2A). Significant increases were also seen for LINEs and short interspersed elements (SINEs), although the magnitude of change was smaller (Table S2). To identify the most affected TEs, we plotted average expression difference versus fold change (Figure 2B), and the 27 elements with the largest expression difference were inspected for autonomous expression using Integrative Genomics Viewer (IGV).29 The most upregulated element was ERV BHIKHARI-2 (bhik-2, also known as crestin), which is transiently expressed during neural crest development, required for this process,30 and re-expressed during melanoma initiation.31 Most of the top upregulated TEs were LTRs/ERVs (79%, 21/27), with bhik-2 ERVs being most affected (71% of LTRs/ERVs, 14/21, solid blue dots in Figure 2B represent bhik-2 elements, example in Figures 2B, 2C, and S2C). Most of these TEs were expressed independently of nearby genes (56%, 15/27), including all 14 bhik-2 elements. To test whether other NEXT substrates were affected by Z18trunc expression, we conducted ribo-minus RNA-seq of Z18trunc (n = 3) versus eGFP (n = 3) melanomas (Figure S2D). PROMPTs and eRNAs did not accumulate in Z18trunc-expressing melanomas (Figure S2E; Table S3). Thus, Z18trunc promotes ERV RNA accumulation without broadly affecting other NEXT targets,3 suggesting selective ERV RNA regulation. Because Z18 is conserved between zebrafish and humans, we asked whether it similarly regulates ERV RNA accumulation in human cell lines. We used SQuIRE to quantify TE expression from publicly available pA-selected RNA-seq from the Cancer Cell Line Encyclopedia (CCLE).32 Cell lines with deleterious Z18 mutations (Z18mut) (n = 11) were compared with matched Z18 wild-type (Z18WT) cell lines (cancer-type, sex, driver mutations) (n = 11) (Figure S2F; Data S2, “CCLE cell lines”). Average log2 fold change in TE expression was plotted for TEs that were significantly different (p < 0.05) between Z18mut and Z18WT cell lines (Figure 2D). All TE classes were upregulated in Z18mut lines relative to controls, with the most regulated elements being LTRs/ERVs (Figure 2D, blue distribution extends to higher values). Using DESeq233 normalization to distribute the data for easier visualization, we found that most of the significantly upregulated TEs were LTRs/ERVs (10/30 upregulated), with seven loci from the HERV-H subfamily (Figure 2E), which manual inspection showed to be derived predominantly from colon cancer lines. We quantified ERV expression and found significant enrichment of LTRs/ERVs in Z18mut relative to Z18WT CCLE cell lines (Figure 2F, left). One significantly upregulated, autonomously expressed ERV at chr16:2659700 was expressed in multiple cancer types, initiated in one element and readthrough several additional annotated TE intervals before terminating (Figure S2G). Thus, Z18-mutant cell lines across cancers upregulate TEs, especially ERVs. Because no Z18-mutant CCLE melanoma cell lines had available RNA-seq, we analyzed TCGA data to assess Z18’s role in ERV regulation in melanoma. We quantified ERV expression in Z18mut and Z18WT TCGA skin cutaneous melanoma (SKCM) with publicly available pA-selected RNA-seq using SQuIRE. Patient samples with deleterious Z18 mutations (alpha missense >0.564, nonsense, or frameshift) were selected (n = 11). Matched Z18WT samples were also selected (stage, sex, age, and driver mutations [BRAF, NRAS, and NF1]) (n = 11). This analysis showed significant accumulation of LTRs/ERVs in Z18mut SKCM patients (Figure 2F, left middle; Data S2, “TCGA SKCM patients”). Because Z18 mutations are the most common in UCEC and COAD cancers, we selected TCGA Z18mut cases: UCEC (n = 14) and COAD (n = 11). We similarly analyzed LTR/ERV expression in the aforementioned datasets and found significant enrichment of LTRs/ERVs in Z18mut relative to Z18WT UCEC patients (Figure 2F, right middle; Data S2, “TCGA UCEC patients”). Global LTR/ERV upregulation was not observed in Z18mut COAD (Figure 2F, right; Data S2, “TCGA COAD patients”), but 50% of the most upregulated TEs in Z18mut COAD were ERVs (15/30), and 53% of these were HERV-H (8/15), the same subfamily found to be upregulated in Z18mut CCLE cell lines, suggesting that Z18 plays a conserved role in suppressing HERV-H elements in colon cancer. Overall, these analyses suggest that Z18-mediated ERV RNA suppression is conserved across species and relevant to multiple cancer types. Z18trunc stabilizes ERV RNA in a dominant-negative manner To test whether Z18trunc promotes ERV RNA accumulation, we generated isogenic clonal human melanoma cell lines that were heterozygous for Z18 patient truncating mutations using CRISPR homology-directed repair (HDR) (Z18trunc/+, n = 3) or Z18WT (n = 2) (Figures 3A and S3A–S3E; Table S4). We recovered one HDR-edited Z18trunc/+ clone (R680Gfs*5) and two clones harboring other Z18trunc/+ patient mutations. To test whether Z18trunc interferes with Z18WT, human melanoma cells with Z18 heterozygous loss-of-function mutations (Z18−/+) were also generated (n = 3) (Figures S3F–S3G). The guide RNA (gRNA) used to make Z18trunc/+ clonal cell lines yielded out-of-frame truncating mutations in 83.3% of edited cell clones (20/24; Figure 3B), whereas the two gRNAs used for heterozygous loss-of-function clones resulted in fewer out-of-frame mutations (10/23, 43.5%; Figure 3C; gRNA locations in Figure S3A), suggesting selection for Z18trunc/+ mutations. Because no viable homozygous loss-of-function clones were recovered (0/23 clones with mutations), we also performed transient bulk CRISPR (Figures S3H and S3I) to assess the Z18 loss-of-function phenotype. To determine if Z18trunc regulates ERV RNA in a dominant-negative manner, we used SQuIRE to quantify TEs from pA-selected RNA-seq in the Z18 allelic series. Z18trunc/+ upregulated all TE classes relative to Z18WT, with LTRs/ERVs most affected (Figure 3D, blue distribution extends to higher values). Bulk Z18 CRISPR-deleted cells also showed upregulation of all TEs, except SINEs (Figure S3J). Given the variability in SINE changes across models, we focused subsequent analyses on ERVs, which showed robust and consistent upregulation across Z18 perturbations. The most upregulated autonomously expressed LTR in the Z18trunc/+ cell lines was chr14:105032869 MER4-int (hereafter MER4-int) (Figures 3E, 3F, and S3K, qPCR primer location; Figure 3F; Table S4). Z18 loss of function via bulk CRISPR deletion also resulted in MER4-int upregulation (Figure 3F, top 6 rows; Figure 3G, left). Notably, MER4-int accumulated to higher levels in Z18trunc/+ cells than in cells with loss of a single Z18 allele (Figure 3F, last six rows), suggesting that Z18trunc interferes with Z18WT-mediated ERV RNA clearance and acts in a dominant-negative manner. Z18trunc/+ likewise increased another significantly regulated and autonomously expressed LTR/ERV, HERVK3-int (chr16:35516910) (hereafter HERVK3-int) (Figures S3L and S3M, qPCR primer location S3M; Table S4). These data support dominant-negative ERV RNA accumulation by Z18trunc/+. To assess the impact of Z18trunc mutations on the broader transcriptome, gene set enrichment analysis (GSEA)34 was performed using differentially expressed RefSeq transcripts obtained from pA-selected bulk RNA-seq of the Z18trunc/+ versus control single-cell clones. Of the Hallmark signatures,35 the most significantly upregulated pathway was “TNFA_SIGNALING_VIA_NFKB” (family-wise error rate p = 0, normalized enrichment score 2.44) (Figure S3N; Data S3, “MSigDB Hallmark Z18trunc vs. Ctrl”), and the two most upregulated pathways from the Chemical and Genetic Perturbation (CGP) datasets were associated with viral infection, suggesting viral mimicry (“RESPIRATORY_SYNCYTIAL_VIRUS_INFECTION_A594_CELLS_UP” and “HUMAN_PARAINFLUENZA_VIRUS_3_INFECTION_A594_CELLS_UP”) (Figure S3O; Data S3, “CPG Z18trunc vs. Ctrl”). We hypothesize that widespread ERV RNA accumulation induces viral mimicry, as reported for ERV transcriptional derepression.17,36,37 Because CRISPR-HDR and single-cell cloning introduce heterogeneity, and because Z18trunc functions in a dominant-negative manner, we generated Z18trunc- and Clover-expressing control human melanoma cell lines (Figure S3P). We quantified TE expression from pA-selected RNA-seq and visualized significantly changed (p < 0.05) loci. Consistent with our previous findings, multiple TE classes were upregulated, with LTRs/ERVs being the most affected (Figure 3G, middle; Figures S3Q and S3R). A union of significantly changed ERVs/LTRs showed that the magnitude of accumulation versus controls was similar (Figure S3S). CDK13-dependent ptRNAs were unaffected (Figure S3T). To measure the effects of Z18trunc on additional NEXT substrates, we performed ribo-minus RNA-seq on Z18trunc- versus Clover-expressing A375s and again found that all TEs were broadly regulated (Figure S3U), with no change in other NEXT targets (Figure S3V). Polyadenylated TEs were more strongly regulated than non-polyadenylated TEs (pA-selected versus rRNA-depleted RNA-seq) (Figure S3W), suggesting that Z18 may preferentially regulate polyadenylated substrates. To test for a functional role in colon cancer, we performed parallel expression of Z18trunc or Clover in SW480 colon cancer cells and found that ERVs/LTRs were significantly upregulated (Figure 3G, right; Figure S3X), although the effect size was smaller, consistent with Z18 playing a more limited role in LTR/ERV regulation in colon cancer. To determine if the increase in the MER4-int transcript was due to defective degradation, we measured its decay kinetics. Using 4sU pulse-chase, we found significant stabilization of MER4-int in Z18trunc-expressing (n = 3) relative to Clover-expressing (n = 3) cell lines (Figures 3H and S3Y). MER4-int was also stabilized in Z18trunc/+ as compared to Z18WT (Figure 3I). To assess MER4-int transcription, nascent RNA38 qPCR was completed for Clover- and Z18trunc-expressing human melanoma cells. Z18trunc cells showed higher MER4-int transcription (Figure 3J), consistent with a report that LINE-1 chromatin is more accessible upon depletion of ZCCHC8, a core NEXT component.9 These data show that the MER4-int transcript is stabilized in Z18trunc-expressing and Z18trunc/+ cells and suggest that MER4-int decay may play a role in its own chromatin silencing. Together, our data are consistent with a model where Z18trunc stabilizes ERV messages in a dominant-negative manner. ZC3H18trunc directly promotes oncogenic ERV RNA stabilization To define the mechanism, we identified proteins that bind full-length (Z18fl) and truncated Z18 via immunoprecipitation mass spectrometry (IP-MS). V5-tagged Z18fl and Z18trunc were transiently expressed in A375 human melanoma cells. Nuclear protein was extracted under native conditions, and Z18 was immunoprecipitated (IPed) in triplicate using a V5 antibody. IPs were performed in the presence or absence of RNase A/T1 to define RNA-dependent and -independent protein interactions (Figure 4A). Data were filtered for proteins that were >4.5× enriched over a published control IP6 and >4 total peptides per replicate in the most permissive condition (Data S4). Z18fl abundantly bound NEXT components, ZCCHC8 and MTR4 (Figure 4B, left), consistent with our data showing that Z18 regulates ERV RNA turnover, which requires NEXT.10 We then compared Z18fl- and Z18trunc-interacting proteins (Figure S4A). As expected, the loss of the C-terminal RNA surveillance recruitment domain caused a significant loss of MTR4 and ZCCHC8 binding to Z18trunc compared with Z18fl (Figures 4B, left; Figure S4B). Four splicing-related proteins also showed reduced binding to Z18trunc (FNBP4, PRPF4B, RBM25, and PRPF40A) (Figure S4C). In contrast, PABPC1 and SRRT had stable or increased binding to Z18trunc (Figures 4B, right; Figure S4D). PABPC1 is a polyadenosine-binding protein that is often amplified in cancer39 and can protect RNAs from decay.40 Although PABPC1 is essential for LINE-1 retrotransposition,41 it has no reported role in NEXT. In contrast, SRRT is a cap-binding protein that associates with NEXT.42 No additional cap-binding proteins were detected above background (NCBP1 [CBP80], NCBP2 [CBP20], NCBP3, PHAX, or eIF4E). Thus, Z18trunc loses NEXT binding but retains PABPC1 and SRRT, consistent with direct ERV RNA stabilization. To consider how Z18trunc exerts its dominant-negative activity, we proposed two models: Z18trunc could bind and sequester (1) Z18WT from its RNA substrates (Figure S4E, model 1) or (2) substrate RNAs away from Z18WT (Figure S4E, model 2). In model 1, Z18trunc should be predominantly localized in the same compartment as Z18fl and should efficiently bind Z18fl. We previously noticed by IP-MS that Z18trunc was found at lower levels in the nuclear fraction than Z18fl (Figure S4F). We hypothesized that Z18trunc was being exported to the cytoplasm, as was previously seen for a C-terminal mutant of Z18.23 V5-tagged Z18 immunoblots indicated that Z18fl was enriched in the nucleus as expected, while Z18trunc was enriched in the cytoplasm (Figure S4G) where it continued to associate with SRRT and PABPC1 (Figure S4H). Z18 immunofluorescence (IF) in Z18trunc-expressing cells confirmed Z18trunc cytoplasmic localization versus control A375 cells (Figures 4C and 4D). Consistently, IF in Z18trunc/+ versus control Z18WT lines showed a higher cytoplasmic to nuclear ratio of Z18 (Figures 4E and S4I). These data indicate that Z18trunc is cytoplasm enriched. To test whether Z18trunc could sequester Z18WT from its substrate, we asked whether Z18trunc bound Z18fl. Z18 IP-MS unique peptide measurements were plotted along the protein length (Figure S4J). As expected, Z18trunc samples (+/− RNase) had only a few peptides measured distal to the truncation site, suggesting that Z18trunc does not pull down Z18fl. To confirm that Z18trunc does not sequester Z18fl, we performed IP-western blot analyses of Z18trunc from nuclear and cytoplasmic fractions to detect Z18fl, finding no signal (arrows in Figures S4K and S4L). Because Z18trunc is dominant negative, does not sequester Z18WT, and is cytoplasm enriched, we hypothesized that it sequesters ERV RNA away from the surveillance machinery. To determine the localization of Z18-regulated ERV RNA, we performed RNA in situ hybridization for the specific MER4-int transcript using BaseScope in Z18trunc-expressing versus control A375 cells. We found that the MER4-int transcript is enriched in the cytoplasm relative to the nucleus in Z18trunc-expressing cell lines (n = 2) versus control cells (Figures 4F, 4G, and S4M–S4O). This suggests that Z18trunc binds and stabilizes ERV messages in the cytoplasm, sequestering them from the degradation machinery. We next considered how Z18trunc might exert its dominant-negative activity by directly binding ERV RNA (Figure S4E, model 2). Since the ability of Z18 to regulate retroelement RNA is highly evolutionarily conserved, we analyzed Z18’s homology to determine how Z18 could bind RNA. By aligning Z18 homologs from human, mouse, zebrafish, C. elegans, and D. melanogaster using a tool called MUSCLE,43 we found two highly conserved domains: the RNA-surveillance recruitment domain and the zinc-finger domain after which Z18 is named (zinc-finger alignment in Figure S4P). Z18 is one of 57 human proteins with a CCCH zinc-finger motif, a protein class that often binds RNA and functions in immunity.44 Since the Z18 structure is unknown, we used the RBM2245 (PDB: 6ID1) structure to create a homology model of the Z18 zinc-finger domain (Figure S4Q). Our model predicted that the CCCH zinc-finger domain has a Zn2+ ion coordinating with C224, C232, C238, and H242 (Figure 4H) and that the highly conserved aromatic residues F226, W234, and F240 directly bind RNA via pi-pi stacking interactions (Figure 4I). We hypothesized that Z18trunc binds RNA via the aromatic amino acids in the zinc-finger domain. To test whether Z18 directly binds ERV RNA via the aromatic amino acids in the zinc-finger domain, we used photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation (PAR-CLIP).46 V5-tagged Clover, Z18fl, Z18trunc, and Z18trunc zinc-finger aromatic mutant (hereafter Z18truncZnFmut) were transiently expressed in A375 human melanoma cells (n = 3 each) and IPed. Bound RNA was sequenced, and differential TE binding relative to Clover was assessed using SQuIRE (Figures 4J and S4R–S4T). Z18fl significantly bound numerous TE RNAs (4,702), Z18trunc maintained binding to most TE RNAs (4,134), while Z18truncZnFmut bound fewer TE RNAs (1,266) (p < 0.05, DESeq2) (Figures 4K and S4U, example transcript in S4V, S4W, and S4X). These data show that aromatic residues in the Z18 zinc-finger domain are required for stable TE binding. Z18trunc bound significantly more LINE elements than Z18fl, so we asked whether LINE-1 retrotransposition contributes to Z18mut phenotypes. Using catalogs of published somatically acquired LINE-1 insertions in a pan-cancer study,47 we found no statistical increase in LINE-1 retrotransposition in Z18mut cancers (average Z18mut 10.09, n = 11; average rest 6.57, n = 326; p = 0.72, two-sided t test). We also found no change in LINE-1 open reading frame 1 (ORF1) protein expression in the Z18trunc-expressing cell lines (Figure S4Y). Thus, Z18fl and Z18trunc directly bind LINE and LTR/ERV RNAs, and zinc-finger aromatic residues are required for stable TE binding. To determine the specific RNA sequences bound by Z18 in the PAR-CLIP data, we used wavClusteR,48 to call high-confidence T-to-C transition regions (clusters). Z18fl and Z18trunc had a similar number of clusters (296,229 and 284,329, respectively) while Clover yielded fewer clusters (144,336), which represented background RNA binding. To determine the identity of Z18-bound RNAs, clusters were annotated for LTRs, LINEs, introns, and exons (Figures S4Z–S4ZC). We found that Z18fl and Z18trunc clusters were enriched for LTRs, LINEs, and introns relative to exons (Figures 4L, 4M, and S4Z–S4ZC; Table S5, example Z18-bound transcript with intron inclusion in Figure S4ZD). In addition, Z18trunc bound other NEXT targets, such as PROMPTs and eRNAs (Figures S4ZE and S4ZF), which were not regulated by Z18trunc (Figure S3V). This is consistent with our findings that Z18trunc preferentially regulates polyadenylated TEs (Figures S3Q versus S3U and S3W) and retains binding to PABPC1, a polyA-binding protein (Figure 4B). Together, these data indicate that Z18trunc binds and regulates polyadenylated TEs, particularly ERVs. To determine which RNA motifs are bound by Z18, a simple enrichment analysis (SEA) was performed.49 This analysis revealed significant enrichment of multiple motifs within a subset of sequences, suggesting that Z18 does not bind to a specific motif, consistent with published reports50 (Table S6). We noticed that many of the most enriched motifs were A/U rich and found that, among the top 12 (log q value <−10, highest true-positive %), A and U made up 83% of the single-nucleotide bases (Figure S4ZG). As introns51 and retroelement52 messages have higher AU nucleotide content than exons, and AU RNA was enriched in the top motifs, we tested whether Z18-bound regions were broadly enriched for AU RNA relative to Clover. We found that both Z18fl and Z18trunc bound more AU-rich RNA than Clover, and this effect was decreased for the Z18truncZnFmut (Figure S4ZH). These analyses indicate that Z18fl and Z18trunc directly bind and regulate AU-rich polyadenylated retrotransposon RNA via conserved aromatic residues in the zinc-finger domain, supporting a dominant-negative mechanism via RNA binding. Z18-regulated ERV RNA is sufficient and necessary for Z18 mutation-associated melanomagenesis Because Z18trunc accelerates zebrafish melanoma and stabilizes ERV RNA, we tested whether ERV RNA directly contributes to oncogenesis. We expressed the two most significantly upregulated ERVs from Z18trunc-expressing zebrafish melanomas, chr4:42592945 bhik-2 (hereafter bhik-2–42) and chr4:37858940 bhik-2 (hereafter bhik-2–37), in the zebrafish system. Notably, expression of either ERV was sufficient to expedite melanoma onset in zebrafish (Figures 5A and S5A–S5C; Table S4). Both zebrafish ERVs are solo LTRs, i.e., they underwent recombination, resulting in loss of intervening protein-coding regions, and are therefore predicted to be noncoding. Thus, we hypothesize that bhik-2 RNA can directly contribute to oncogenesis. To test whether Z18-regulated ERV-driven oncogenesis is evolutionarily conserved, we expressed the two ERV RNAs that are most stabilized in Z18trunc/+ and Z18trunc-expressing human melanoma cells, MER4-int RNA and HERVK3-int. Both human ERVs have spliced and unspliced isoforms, so we tested the oncogenic potential of both transcripts. We expressed spliced MER4-int (MER4-int exon-exon) and unspliced MER4-int (MER4-int exon-intron) (Figure S5D) in the triple-mutant zebrafish melanoma model. MER4-int exon-intron expression produced increased black patches (early melanoma) at 10.7 weeks (Figure 5B, left; Figure S5E) and expedited melanoma onset compared to MER4-int exon-exon and eGFP expression (Figure 5B, right; Figures S5I and S5J). We also tested whether Z18trunc-regulated HERVK3-int would expedite melanoma (unspliced: HERVK3-int exon-intron; spliced: HERVK3-int exon-exon) (Figure S5F) and found that both transcripts caused increased black patches at 10.7 weeks (Figures S5G and S5H, left) and expedited melanoma onset compared to an eGFP-expressing control (Figures S5H, right, S5K, and S5L). These data indicate that human ERV RNA itself can contribute to oncogenesis in a conserved manner. To test whether ERV RNA is required for Z18 mutant-driven oncogenesis in zebrafish, we developed a melanocyte-specific CRISPR inhibition (CRISPRi) approach to inhibit bhik-2 expression. Catalytically dead Cas9 fused to a transcriptional repressor (dCas9-KRAB-MeCP2) was expressed in melanocytes using the mitfa promoter; gRNAs were expressed using a ubiquitous promoter, and a Mitfa minigene driven by the mitfa promoter enabled cell-autonomous, melanocyte-specific CRISPRi (Figure S5M). We tested the system using a previously validated gRNA targeting tyrosinase, a gene required for pigment production53 (gRNAs in Table S4). CRISPRi of tyrosinase significantly reduced melanin production in melanocytes at 6 days post-fertilization (dpf) in wild-type zebrafish (Figure S5N), indicating functionality of the system. To inhibit bhik-2 expression, we designed gRNAs that are predicted to target multiple bhik-2 elements (13/15 Z18-regulated bhik-2 ERVs) (Figure S5O) and tested the efficiency of injected bhik-2 gRNA and Cas9 protein in whole embryos using a T7E1 assay (Figure S5P). The two most efficient gRNAs were integrated into the melanocyte-specific CRISPRi system, and qPCR in whole embryos confirmed bhik-2 inhibition (Figure S5Q), validating CRISPRi-mediated repression of bhik-2. To test whether bhik-2 expression is required for Z18-mediated oncogenesis, we performed melanocyte-specific bhik-2 CRISPRi versus control in Z18trunc or eGFP zebrafish (schematic of Figure 5C). In Z18trunc-expressing melanocytes, bhik-2 CRISPRi significantly reduced the melanocyte area at 6 dpf (Figures 5D and 5E), while bhik-2 CRISPRi had no effect in eGFP-expressing zebrafish (Figures 5F and 5G). We also found that bhik-2 CRISPRi significantly slowed tumor onset in Z18trunc-expressing melanoma (Figure 5H), while bhik-2 CRISPRi had a minimal effect on tumor onset in eGFP-expressing zebrafish (Figure 5I). Thus, bhik-2 is required for melanocyte development and melanoma onset in the Z18trunc context. To test whether Z18trunc human melanoma cells require ERV RNA, we performed CRISPRi of MER4-int in Z18trunc-expressing and control human melanoma cells. Two gRNAs were confirmed to inhibit MER4-int expression using transient transfection as compared to no transfection or a control gRNA (Figure S5R) (both MER4-int gRNAs, zero perfect genome matches). These two gRNAs were then introduced into Z18trunc- or Clover-expressing A375 human melanoma cell lines using lentivirus (n = 3, each). Cells were selected with antibiotics to ensure CRISPRi delivery, expanded, and stained with crystal violet to assess the cell number. MER4-int gRNAs significantly reduced numbers compared to a control gRNA in both Z18trunc and Clover cell lines (Figures 5J, 5K, and S5S). This result shows that ERV RNA is required for A375 cancer cell growth and supports a functional oncogenic role for ERV RNA in human melanoma. DISCUSSION Our findings identify Z18 mutation as a mechanism of ERV RNA accumulation in cancer and support a direct oncogenic role for ERV RNA. Although we initially expected Z18trunc to affect ptRNAs, it instead caused selective accumulation of retroelement RNAs, especially ERVs, across zebrafish, patient tumors, and engineered human cells. These results broaden our prior work on oncogenic RNA surveillance defects6 and suggest that distinct perturbations in nuclear RNA quality control can promote cancer through accumulation of different aberrant polyadenylated RNAs. We found that Z18 deletions are common and significantly enriched in cancer (36.6% all cancers; 30.0% melanoma). Although Z18’s role in cancer has not been investigated previously, one case report described transformation of severe congenital neutropenia into acute myeloid leukemia, with the offending clone harboring a Z18trunc mutation (777fs),54 suggesting that although rare, truncating mutations are functional. Z18 truncating mutations are statistically enriched just upstream of the domain required to recruit nuclear RNA degradation machinery, supporting a role beyond loss of function. Consistent with this, Z18trunc/+ enhanced ERV RNA accumulation more than loss of one Z18 copy, demonstrating dominant-negative activity. We previously showed that CDK13 mutations also act in a dominant-negative manner,6 suggesting a common mechanism in nuclear RNA surveillance mutations. We hypothesize that dominant-negative mutations perturb RNA metabolism while avoiding lethality from complete loss of function. Z18, which is named for its CCCH zinc-finger domain (ZC3H18), is one of nearly 60 human CCCH zinc-finger proteins.44 Where studied, CCCH zinc fingers bind RNA, and many, including Z18, function in immune responses.44,55 We found that Z18 normally binds ERV RNA and targets it for degradation. Because retroelements comprise nearly half of the human genome,11,56 even low-level transcription could plausibly affect cellular phenotypes.12 Multiple organisms have therefore evolved mechanisms to suppress ERV expression, even for immobile elements. The best described is transcriptional silencing by Krab zinc-finger proteins (KZFPs). The role of KZFPs in limiting TE expression is so profound that multiple studies have shown that TE evolution drives host zinc-finger evolution.57–62 Our work indicates that RNA-binding zinc fingers can promote ERV RNA degradation, complementing DNA-binding zinc-finger-mediated transcriptional repression. The human silencing hub (HUSH) complex has also been shown to repress expression of intron-less retroelements.63 Although ZCCHC8, a component of the NEXT complex, directly interacts with HUSH,10 Z18 was not detected with key HUSH components in prior work23 or in our IP-MS experiments (except PPHLN, below our filtering threshold). Instead, we observed enrichment of splicing-related proteins and found that Z18 tends to bind intron-containing transcripts. For example, two of the most regulated ERV-containing transcripts, MER4-int and HERVK3-int, were spliced with retroelement or host sequences. Spliced ERV RNAs, which are not regulated by HUSH silencing, may therefore require NEXT targeting via Z18, potentially explaining Z18 specificity for TE targets over unspliced RNAs such as PROMPTs and eRNAs. Retroelements, especially LTRs/ERVs,16,17 are upregulated in many cancers, suggesting positive selection for RNA production from TE loci. Intriguingly, proper ERV RNA levels are required during early embryogenesis, with both upregulation64 and downregulation65 causing lethality. Recently, proteins encoded by zebrafish Z18-regulated bhik-2 elements were shown to be required for neural crest development.66 Because cancers often co-opt developmental programs, these studies suggest that, despite their immobility, ERV RNAs may be repurposed to promote oncogenesis. Consistent with this model, zebrafish and human Z18-target ERVs accelerated melanoma onset in zebrafish, showing that aberrant ERV expression itself can contribute to oncogenesis. Z18-regulated bhik-2 ERV RNA was also required for Z18-driven melanoma in zebrafish, and Z18-regulated ERV expression was required for proliferation of human melanoma cells. In clear cell renal cell carcinoma, VHL loss drives HIF2-dependent ERV RNA upregulation despite immune responses to ERV antigens,67 suggesting that other common cancer alterations may likewise activate ERV RNA. Our findings suggest that the Z18-ERV RNA axis may have therapeutic implications. We observed that Z18 mutations activated viral mimicry gene expression, similar to ERV de-repression,17,36,37 which can increase immune infiltration, antigenicity, and responses to immune therapy in multiple cancers.17–19 ERV-encoded antigens are increasingly recognized as targets of anti-tumor immunity.67 We therefore hypothesize that patients with Z18-mutant cancers may respond better to immunotherapy due to activation of innate immune signaling or recognition of translated ERV-encoded peptides. In addition, we hypothesize that transient Z18 inhibition in Z18WT cancers could potentially sensitize tumors to immunotherapy. Given that ~30% of patients harbor Z18 loss or mutation and that Z18 modulation may be therapeutically tractable, this pathway is likely to be clinically relevant. More broadly, our work supports the concept that defective nuclear RNA degradation shapes the cancer transcriptome, contributes to oncogenesis, and reveals therapeutic vulnerabilities. Limitations of the study Our data support an oncogenic role for ERV RNA, but the mechanism remains unknown. We also do not yet fully understand why Z18 mutation or loss selectively affects retroelements while leaving other NEXT substrates intact. In addition, we found that human cells depend on one Z18-regulated ERV RNA, although it remains unclear whether this reflects a broader dependence on ERV RNA. Finally, because the ERV RNA CRISPRi experiments were performed in a single parental human melanoma cell line, additional human cell models are needed to establish generalizability. RESOURCE AVAILABILITY Lead contact Requests for further information and resources will be fulfilled by the lead contact, Megan L. Insco (megan_insco@dfci.harvard.edu). Materials availability All zebrafish strains, human cell lines, and DNA vectors are available through the corresponding author. The zebrafish melanocyte-specific CRISPRi vector has been deposited (AddGene, 251742). Data and code availability All generated data have been deposited at GEO and are publicly available as of the data of publication. Please see the key resources table for details. All generated data have been deposited at GEO and are publicly available as of the data of publication. Please see the key resources table for details. This paper also analyzes publicly available data from GDC and GEO. Details are provided in STAR Methods. This paper also analyzes publicly available data from GDC and GEO. Details are provided in STAR Methods. This paper does not report original code. This paper does not report original code. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. STAR★METHODS EXPERIMENTAL MODEL DETAILS Animals Animal studies were approved by the Animal Care and Use Committee at the Dana-Farber Cancer Institute (IACUC 21–027). Three strains of zebrafish were used: wild-type AB, Casper (mitfa−/−; roy−/−),68 and the Triple melanoma model (mitfa−/−; mitfa:BRAFV600E; p53−/−).27 Casper fish were genotyped visually because they have a distinctive appearance resulting from the absence of most pigmentation cells. The Triple melanoma model was outcrossed to maintain strain health and then incrossed every four generations; incrossed animals were genotyped to confirm homozygosity for the three genetic alterations. The developmental stage varied by experiment and is described in the relevant sections below. Animals were fed and monitored daily for changes in health or signs of distress as specified in the animal protocol. Water parameters were continuously monitored. Animals exhibiting signs of distress or tumors exceeding 10% of adult zebrafish body weight were isolated and euthanized in ice water. Matching animal inclusion by sex was not feasible, as sex is environmentally determined and not morphologically apparent until 3 months of age. Age and sex were matched for tumor collection where possible. Cell lines A375 human melanoma cells (ATCC CRL-1619) or SW480 colon adenocarcinoma cells (ATCC CCL-228) were identity-verified via short tandem repeat analysis, then used for transient transfections or stable line generation. Routine monthly mycoplasma testing was performed (Sigma, MP0025). All cell lines were mycoplasma negative. Cells were grown at 37°C in DMEM with 4.5 g/L D-glucose and L-glutamine supplementation with 10% FBS and penicillin/streptomycin or selection antibiotics as needed. A375s are derived from a female and SW480 cells are derived from a male; we do not expect sex to affect results, as Z18-mutant-derived ERV expression occurs in both male and female patients. METHOD DETAILS Mutational analysis of patient samples We downloaded cutaneous melanoma datasets consisting of 1,347 samples from cBioPortal39 and the International Cancer Genome Consortium (ICGC) Data Portal (Data S1, “Z18 Mut & Melanoma Patient Refs”) to identify all Z18 mutations. Mutations that matched germline mutations were excluded. The PanCancer dataset20 was obtained from The Cancer Genome Atlas (TCGA) and underwent PolyPhen-221 analysis (Data S1, “PolyPhen-2 Results”). GISTIC statistics were gathered from tumorscape.org via the “Gene-Centric” analysis.22 Next, a dataset of nonredundant studies spanning various cancer types from cBioPortal was compiled. In total, 85 unique studies were identified (Data S1, “Pan-cancer Patient Sample Refs”), encompassing 23,411 samples, each containing at least one sample with a mutation in the Z18 gene. Duplicate or serial samples from the same individual were excluded. Amino acid residues with >5 mutations were plotted. To analyze whether truncating mutations are enriched upstream of the RNA surveillance recruitment domain, a Fisher’s exact test was used (Data S1, “Z18trunc Fisher’s Exact Test”). MSIsensor83 scores were used to identify microsatellite-unstable samples within the TCGA PanCancer dataset.20 Samples with Z18 truncating and nontruncating mutations were compared to Z18 wild-type (Z18WT) samples to determine which were associated with microsatellite instability (MSIsensor >10) using an odds ratio and a Fisher’s exact test (two-sided) (Table S1). TCGA Z18 mutant repeat element expression analysis Melanoma (SKCM) sample selection Samples with deleterious Z18 missense (AlphaMissense >0.564), nonsense, or frameshift mutations with publicly available RNA-seq and clinical information were downloaded on 4/30/25 (TCGA Firehouse Legacy and DFCI 2019)20,84 (n = 10). Age, sex, stage, and driver mutations (BRAF [V600 or type II BRAF], NRAS Q61, and NF1) were also downloaded. Z18WT samples (n = 10) were selected to match the Z18mut samples by study, stage, sex, and driver mutation and were then downloaded. Age was only available for TCGA samples and was matched accordingly. See also Data S2. Uterine cancer (UCEC) sample selection Samples with Z18 missense (AlphaMissense >0.564), nonsense, or frameshift mutations with publicly available RNA-seq and clinical information were downloaded on 8/10/2520 (n = 14). Samples with Z18 hotspot truncations (amino acid 679–901) were prioritized. Sex, age, and driver mutations that are reported to affect >20% of patients, as well as the three most frequently mutated mismatch repair genes were downloaded (PTEN, PIK3CA, CTNNB1, ARID1A, KRAS, TP53, PIK3R1, ZC3H18, MLH1, MSH2, MSH6). Z18WT samples (n = 14) were selected to match the Z18mut samples by sex and driver mutation and were then downloaded. Stage was not considered, as staging information was not available. See also Data S2. Colon cancer (COAD) sample selection Samples with Z18 missense (AlphaMissense >0.564), nonsense, or frameshift mutations with publicly available RNA-seq and clinical information were downloaded on 8/10/2520 (n = 11). Samples with Z18 hotspot truncations (aa 679–901) were prioritized. Stage, sex, age, and driver mutations that are reported to affect >20% of patients, as well as the three most frequently mutated mismatch repair genes were downloaded (KRAS, APC, TP53, PIK3CA, ZC3H18, MLH1, MSH2, MSH6). Z18WT samples (n = 11) were selected to match the Z18mut samples by stage, sex, age, and driver mutations, and were then downloaded. See also Data S2. Analysis BAM files pertaining to the above samples were converted to FASTQ prior to processing with SQuIRE. LTR/ERV loci-specific FPKM values were collected. All “NA” FPKM values were replaced with 0s (to account for no detected TE expression) and only TE loci expressed in at least one sample on the same strand as the TE locus were included. Wilcoxon tests were used to compare expression of individual LTR/ERV locus in mutant versus wild-type samples and significantly changed loci (p < 0.05) were plotted for all three TCGA datasets. Zebrafish experiments Cloning The MiniCoopR (MCR) Gateway system was used.27 MCR overexpression constructs were assembled using Gateway vectors including the mitfa promoter, relevant coding sequence, 3′ polyA tail, and MCR destination vector with LR Clonase II Plus (Thermofisher, 12538120). The human Z18 coding sequence was obtained from Dharmacon (Accession: MHS6278–202759301 Clone ID: 6042036). Truncated Z18 was cloned with the R680Gfs*5 (hereafter Z18trunc) patient mutation using in vitro mutagenesis (NEB, E0554S). Gene fragments for the chr4:42592945 BHIKHARI-2 ERV (hereafter bhik-2–42) and chr4:37858940 BHIKHARI-2 ERV (hereafter bhik-2–37) were ordered from BioTwist, and each fragment was assembled into SalI and EcoRI linearized MCR vector (NEB, E2621). The MER4-int (exon-exon and exon-intron) and HERVK3-int (exon-exon and exon-intron) constructs were synthesized into the MCR MCS vector by GenScript. Melanoma model ARRIVE guidelines were followed where possible.85 Experiments were performed as published.27,31 Briefly, p53−/−; mitfa:BRAFV600E; mitfa−/− (triple-mutant) one-cell embryos were injected with either 20 ng/μL of control or experimental DNA along with in vitro-transcribed Tol2 RNA for integration.27 In all experiments, DNA that overexpressed the gene of interest was marked with a mitfa minigene to rescue Mitfa, allowing cell-autonomous melanocyte genetics (schematic in Figure S1A). Control vectors expressed eGFP. When multiple vectors were injected, total DNA was 20 ng/μL, split equally among vectors. Embryos were sorted for melanocyte rescue (vector integration) at 6 days post-fertilization (dpf). Equal numbers of zebrafish were raised in control and experimental tanks to control for density effects. Zebrafish with >10% of body length showing melanocyte rescue were anesthetized in tricaine and imaged with the Nikon SMZ18 stereo microscope at 9 weeks (for Z18trunc) or 10.7 weeks (for ERV-expressing). Black patches that disrupted normal stripes and had a diameter greater than the distance between that zebrafish’s eyes were quantified from fish images. Zebrafish were scored for emergence of raised melanoma lesions ≥1 mm2 in subsequent weeks. Melanoma-free survival curves and Log rank tests were generated in Prism. Melanocyte-specific CRISPRi Zebrafish codon-optimized dCas9-KRAB-MeCP2 (CRISPRi) sequence was provided by the McCallion lab53 and synthesized into the MCR expression vector by GenScript. Melanocyte-specific CRISPRi function was validated using a gRNA targeting the tyrosinase promoter53 in wild-type zebrafish (using Tol2 RNA for integration). Images to assess melanocyte pigment intensity were then acquired at 6 dpf. gRNAs 1 and 2 targeting bhik-2 were selected via CRISPOR74 and Integrated DNA Technologies (IDT), then cloned into the CRISPRi MCR vector. Both selected gRNAs were predicted to target 13 of the15 top Z18-regulated bhik-2 elements. gRNA efficiency was assessed by injecting 10ng of in vitro-transcribed bhik-2 gRNA +200ng of Cas9 protein (PNA Bio, CP01) into mitfa−/−; roy−/− (Casper) embryos.68 Genomic DNA was collected at 1 day post-fertilization (dpf), PCR-amplified across the bhik-2–37 gRNA cut site, and analyzed by T7E1 assay. Both gRNAs cut efficiently and were used in subsequent experiments; a no-gRNA CRISPRi vector served as the control. Melanocyte-specific CRISPRi gRNA efficiency was tested at 3 dpf using bhik-2–42 qPCR in whole triple-mutant embryos. In triple-mutant embryos, CRISPRi control, bhik-2 gRNA1, or gRNA2 were co-injected with either eGFP or Z18trunc vector, and melanocyte area was quantified at 6 dpf using ImageJ ROI Manager. Melanoma scoring, plotting, and analyses were as described above. Primers in Table S4. Western blotting Protein was quantified using the DC Protein Assay (Bio-Rad, 5000116). The following antibodies were used: anti-Z18 (Novus, NBP1–82570, 1:1000), anti-beta actin (CST, 3700S, 1:1000), anti-GFP (Santa Cruz, sc-9996, 1:1000). Rabbit or mouse secondary HRP antibodies were incubated for 1 h at room temperature (20°C) (CST, 7074S or 7076S, 1:2000). Films were developed with Pierce ECL substrate (Thermofisher, 32106). RNA-seq Triple-mutant zebrafish melanomas from different individual fish with expression of eGFP (n = 3) or Z18trunc (n = 3) were collected. Tumors were homogenized, RNA was purified (Zymo Research, R1054), DNA was removed, and RNA was polyA-selected (NEB, E7490S). Additional eGFP (n = 3) or Z18trunc (n = 3) melanomas were processed similarly and rRNA-depleted (NEB, E7405). All samples underwent stranded library preparation (NEB, E7760S) and were sequenced on a HiSeq 4000. RNA-seq custom genome mapping to verify transgenes The raw sequence reads from polyA-selected and ribo-minus RNA-seq were trimmed using Cutadapt (v. 4.6) with the parameters: cutadapt -j 4 -q 30 -a GATCGGAAGAGCACACGTCTGAACTCCAGTCAC -a CTGTCTCTTATACACATCT -g AGATGTGTATAAGAGACAG -A GATCGGAAGAGCGTCGTGTAGGGAAAGAGTGTAGATCTCGGTGGTCGCCGTATCATT -A CTGTCTCTTATACACATCT -G AGATGTGTATAAGAGACAG -m 50 -e 0.02.86 The trimmed reads were mapped to a custom reference genome containing the chromosomes of the Danio rerio genome build danRer11 plus the human gene Z18. Mapping was performed using STAR (v. 2.5.3a) in paired-end mode with parameters –outSAMtype BAM SortedByCoordinate –quantMode GeneCounts –outWigStrand Stranded –outWigType bedGraph.80 bedGraph files were converted to bigWig format using bedGraphToBigWig.70 bigWig files were visualized in Integrative Genomics Viewer (IGV).77 RNA-seq transposable elements (TE) analysis TEs were measured from polyA-selected RNA-seq with SQuIRE28, which uses stringent mapping coupled with an expectation maximization-based algorithm. Reads were mapped to the danRer11 genome, counted, and examined for differential expression of individual TE loci as annotated by the RepeatMasker GTF (downloaded from UCSC on 9/20/2022). Average expression of each TE locus was calculated for all samples. Only loci with expression on the correct strand in at least one sample were considered. In cases where an individual locus was expressed in a subset of samples, “NaN” values were updated to “0”. A Wilcoxon signed-rank test with continuity correction was used to determine statistical differences in TE expression between Z18trunc versus eGFP conditions. Mean LTR/ERV loci expression (FPKM) was visualized in Figure 2A. Statistics for other TE classes are included in Table S2. The average FPKM log2 fold change was plotted by the difference for all loci in Figure 2B. Uniquely mapping reads for the most highly expressed TEs with a significant change were visualized in IGV.29 RNA-seq analysis of canonical NEXT targets Ribo-minus RNA-seq reads from eGFP or Z18trunc-expressing zebrafish melanomas were mapped to the danRer11 genome build using STAR 2.5.3a80 with the command parameters: STAR –runMode alignReads –twopassMode Basic –outFilterMultimapNmax 20 –outFilterMismatchNmax 999 –outFIlterMismatchNoverLmax 0.033 –alignIntronMin 70 –alignIntronMax 500000 –alignMatesGapMax 500000 –alignSJoverhangMin 8 –alignSJDBoverhangMin 1 –outSAMstrandField intronMotif –outFilterType BySJout. Bed files containing the genomic coordinates of known NEXT target RNAs were generated as follows. Enhancer regions were identified using a published H3K27ac ChIP-seq bed file from a zebrafish melanoma cell line (GEO: GSM1953851_CKK2_11_peaks.narrowPeak.gz).31 As this was a narrow peak file mapped to Zv9, liftOver from UCSC78 was used to convert Zv9 coordinates to danRer11. Using the intersectBed –v function from Bedtools version 2.30.0,71 H3K27ac-enriched regions that excluded gene bodies were identified as intergenic enhancers, while H3K27ac-enriched regions that overlapped the gene body excluding the 200 nucleotides downstream of the transcription site (to eliminate promoters) were identified as intragenic enhancers. PROMPTs were identified as the 200 nucleotides upstream in the antisense direction of all annotated gene transcription start sites. The mapped reads falling within each region of the NEXT target bed files were counted using the featureCounts function of Subread version 2.0.3.76 Counts were input into DESeq233 for eGFP- versus Z18trunc-expressing samples. RNA-seq analysis of ptRNA targets We obtained the ptRNA genome annotation file6 and evaluated differential exon and ptRNA expression using DEXSeq.75 Human cell line experiments CCLE TE RNA-seq analysis Eleven cell lines from the Cancer Cell Line Encyclopedia (CCLE)87 with Z18 mutations with available RNA-seq data were downloaded corresponding to 4 endometrial/uterine cancer lines, 3 colon/colorectal cancer lines, 2 lung cancer lines, 1 gastric cancer line, and one lymphoma line (Data S2, “CCLE cell lines”) (SRA: PRJNA523380). As controls, we selected 11 samples without Z18 alterations, matched as closely as possible by tumor type, sex, and genetic drivers (putative driver mutations, Deep Deletions, and homdel_rec mutations from OncoPrint downloaded from cBioPortal on 08/02/25). SQuIRE28 was run on FASTQ files. As these data are unstranded, strandedness was not considered. TE loci that were expressed in at least one sample were included in downstream analysis. “NA” values were replaced with 0s. Significantly changed TEs between Z18mut versus Z18WT were included (p < 0.05, paired Wilcoxon test). Average TE loci FPKM expression was computed for Z18WT and Z18mut samples, and log2 fold changes were calculated using a 0.01 pseudocount, stratified by TE class (Figure 2D). SQuIRE call (DESeq2) was used to compare Z18mut and Z18WT cell lines (Figure 2E). Human cell line genetic model generation Z18 allelic series cell lines. gRNAs were selected via CHOPCHOP1473 and cloned into Gecko2.0 vectors (Addgene, 49535) which express both Cas9 and the relevant gRNA in one plasmid (Table S4). gRNA efficiency was assessed via bulk transient transfection (Invitrogen, 100022050) in A375 cells, 72-h puromycin selection, and Z18 immunoblotting. The most efficient gRNAs (>50% protein reduction) were selected for single-cell cloning using limiting dilution. To introduce the Z18trunc/+ mutation, we used CRISPR-mediated homology-directed repair by co-transfecting a vector expressing a sgRNA targeting within 20bp of the 680 mutation site and a single-stranded HDR donor DNA template (Integrated DNA Technologies). Z18trunc/+ single-cell clones were screened by Sanger sequencing, and expression of the truncated protein isoform was confirmed by immunoblotting. Heterozygous loss-of-function (Z18−/+) clones were selected that had (1) Sanger sequencing consistent with loss of one Z18 allele and (2) ~50% protein reduction by immunoblot. Wild-type Z18 (Z18WT) control cell lines were generated from cells transfected with Cas9-expressing vectors with no gRNA. These lines were also single-cell cloned, and intact Z18 was confirmed by Sanger sequencing and immunoblotting. Bulk Z18 CRISPR. Since Z18 null clones were not isolated as expected (Z18 is a common essential gene in DepMap, https://depmap.org/portal88), transient bulk CRISPR was performed using the most efficient guide (gRNA 1, Figure S3A; Table S4) in triplicate. Cells transfected with a Cas9-expressing vector without a gRNA served as controls. Transfected cells were selected with puromycin for 48 h, and Z18 protein reduction was confirmed by immunoblot. Z18trunc-expressing lines. pLENTI CMV Clover (fluorescent control) and pLENTI CMV Z18trunc-expressing human melanoma A375 or SW480 colon cancer cell lines were generated via lentiviral transduction and stable puromycin selection. Lines were made in biological triplicate and maintained in selection antibiotics. RNA-seq RNA was isolated, and genomic DNA was removed from Z18WT (n = 2), Z18trunc/+ (n = 3), and Z18−/+ (n = 3) A375 single-cell clones; bulk-CRISPR control (n = 3) and Z18 CRISPR A375 cells (n = 3); Clover-expressing (n = 3) and Z18trunc-expressing (n = 3) A375 cells; and Clover-expressing (n = 3) and Z18trunc-expressing (n = 3) SW480 cells as described above for zebrafish. Samples were either polyA-selected (NEB, E7490) or rRNA-depleted (NEB, E7405) prior to stranded library preparation (NEB, E7760S). Samples were sequenced on a HiSeq 4000 with paired-end sequencing. RNA-seq TE quantification SQuIRE28 was run on FASTQ files as described above for zebrafish TE quantification, except that data were mapped to hg38. Log2 fold changes in TE FPKM for Z18trunc/+ versus Z18WT that were identified as significant (p < 0.05) using the DESeq233 (SQuIRE) were plotted by TE class (Figure 3D) and by p-value (Figure 3E). GSEA34 was performed using the RefSeq DESeq233 output from SQuIRE. FPKM counts on the same strand as LTR/ERV loci that were expressed in at least one sample were included in downstream analysis. “NA” values were then replaced with 0s. Significantly changed TEs in Z18mut versus Z18WT were filtered (p < 0.05, Wilcoxon test). Z18mut and Z18WT average FPKM values were plotted. A Wilcoxon rank-sum test was used to assess changes between conditions. RNA-seq analysis of canonical NEXT targets Coordinates for promoter-upstream transcripts (PROMPTs), intergenic eRNAs, and intragenic eRNAs were defined using the GENCODE v47 human annotation and H3K27ac ChIP-seq data (ENCODE: ENCSR707BXM). Significant H3K27ac peaks were identified using MACS2. PROMPTs were defined as the antisense 200 bp regions upstream of transcription start sites (TSSs). Intergenic eRNAs were defined as H3K27ac peaks located outside of gene bodies and their 200 bp upstream promoter region. Intragenic eRNAs were defined as H3K27ac peaks within gene bodies but outside of exons and the 200 bp region downstream of the TSS. Read counts for these regions were quantified using featureCounts. Differential expression analysis was then performed using the DESeq2. RNA-seq analysis of ptRNA targets We obtained the ptRNA genome annotation file.6 We evaluated differential exon and ptRNA expression using DEXSeq. For visualization, exons with log2 fold change >8 were considered outliers and excluded from plots. qPCR RNA was isolated, and genomic DNA was removed (Zymo Research, 2052). First-strand cDNA was synthesized (Takara Bio, RR036A) and used for qPCR (Biotium, 31003). RNA expression was plotted relative to GAPDH for human cell lines or gapdh for zebrafish samples and normalized to the control condition. Primers are listed in Table S4. 4sU pulse-chase RNA decay measurement Cell lines were labeled with 50 μM 4-thiouridine (Sigma Aldrich, T4509) for 2 h followed by replacement with 4sU-free media. Cells were harvested immediately (time 0) or at 2 and/or 4 h post media replacement. Total RNA was isolated using TRIzol and treated with DNase I (Thermofisher, 18068015). 10 μg of 4sU-labeled RNA was biotinylated, purified, captured by streptavidin magnetic beads (Invitrogen, 65002), eluted using 100 mM dithiothreitol (DTT), and purified by ethanol precipitation. cDNA was prepared and quantified by qPCR as above. Nascent RNA qPCR A TT-seq protocol38 was used with a qPCR readout. Clover- and Z18trunc-expressing cell lines were labeled with 500 μM 4-thiouridine (Sigma Aldrich, T4509) for 5 min (min), lysed using TRIzol, then RNA was extracted. 50 μg of total RNA was DNase I-treated (Thermofisher, 18068015) and fragmented to 1000–1200 bp fragments with a Covaris E220 ultrasonicator (50 s at 105 peak incident power). Nascent RNA was biotinylated, purified, captured by streptavidin magnetic beads (Invitrogen, 65002), eluted using 100 mM DTT, and purified by ethanol precipitation. cDNA was prepared and quantified by qPCR as above. Western blotting 5–15 μg of protein was run on 10% Tris-glycine gels followed by either semi-dry or wet (30 V, overnight) transfer to PVDF or nitrocellulose membranes. Membranes were blocked with 5% skim milk in TBST. The following primary antibodies were used with a 1:1000 dilution: anti-Z18 (Novus, NBP1–82570), anti-ZCCHC8 (Proteintech, 23374–1-AP), anti-PABPC1 (abcam, ab21060), anti-SRRT (Invitrogen, PA5–31593), anti-GAPDH (CST, 2118), anti-VCL (Sigma, HPA002131), and anti-beta-actin (CST, 3700). These antibodies were used with a 1:2000 dilution: Anti-V5 (abcam, ab27671) and anti-MTR4 (abcam, ab70551). Anti-LINE1 ORF1p (EMD Millipore, clone 4H1) was used with 1:5000 dilution. Rabbit or mouse secondary antibodies were diluted 1:2000 and incubated for 1 h at room temperature (20°C) (HRP: CST 7074S or 7076S) (fluorescent: LICOR 926–32211 or 926–68070). When visualizing immunoprecipitated protein, TidyBlot HRP (Biorad, STAR209P) was used at a 1:400 dilution. Films were developed with ECL substrate (Thermofisher, 32106 or Cytivia, RPN2232). IP-MS Coding sequences of full-length Z18 (Z18fl) and Z18trunc (R680Gfs*5) (as described above in Zebrafish cloning) were assembled using the Gateway cloning system into pcDNA3.2 C-terminal V5 tag destination vectors (Thermo Fisher, 12489019). Each construct was transiently transfected into A375 human melanoma cells in triplicate (Thermo Fisher, L3000008). After 48 h, nuclei were isolated and lysed (Thermo Fisher, 78833). Anti-V5 (V5–10, Sigma, V8012) was conjugated to protein G beads (Thermo Fisher, 10004D) and used to IP V5-tagged proteins in the presence or absence of RNase A/T1 (100 μg/mL, Thermo Fisher, EN0551). IPed proteins were eluted by boiling and submitted for mass spectrometry at the Taplin Mass Spectrometry Facility at Harvard University. Data were filtered for proteins with >4 total peptides in all replicates in the most permissive condition (Z18fl -RNase) and >4.5 average peptide count relative to a previous control IP (Clover-V5)6 (n = 240 proteins). The % of total peptides relative to bait Z18 was calculated. To identify significant changes in direct interactors, a two-sided t test was performed between +RNase Z18fl and Z18trunc conditions. Proteins that were significantly changed between +RNase Z18fl and Z18trunc (p < 0.05) and had an average > 5% of bait in at least one +RNase condition were plotted in GraphPad Prism (n = 8). Immunofluorescence Cells were seeded and grown for 24 h in chamber slides (Thermo Scientific, 154917PK) to ~50% confluence, rinsed with 1x PBS, fixed with 4% paraformaldehyde for 15 min at room temperature, washed three times with 1x PBS, permeabilized with 1x PBS with 0.1% Triton X-100 for 15 min, and blocked with 1x PBS containing 10% normal goat serum for 1 h. Cells were incubated with anti-Z18 (Novus, NBP1–82570) at 1:400 overnight at 4°C. Cells were then washed 3x in 1x PBS, incubated with Alexa 488-conjugated anti-rabbit secondary antibody (Invitrogen, A11008) at 1:2000 for 1 h at room temperature in the dark. Cells were washed in 1x PBS, counterstained with DAPI, and mounted. Images were acquired using a Nikon ECLIPSE TS2 inverted microscope with identical settings for all conditions. Raw images were imported into Fiji for quantification. BaseScope RNA in situ hybridization BaseScope was performed using the BaseScope Detection Reagent Kit-RED user manual provided by Advanced Cell Diagnostics (ACD). Cells were seeded and grown for 24 h in chamber slides (Thermo Scientific, 154917PK) to ~50% confluence, washed once with 1x PBS, fixed with 10% neutral-buffered formalin for 30 min, and then pretreated with protease III for 30 min at 40°C as supplied by ACD. Slides were washed twice with 1x PBS, incubated with MER4-int probe (from ACD) for 2 h, and then incubated sequentially with AMP1–8 reagents at 40°C in a HybEZ oven. Slides were incubated with Fast Red for 10 min at room temperature in the dark, stained with DAPI, mounted, and images were acquired with a Zeiss LSM 980 confocal microscope with 0.98 μm intervals. We used a Cy3 filter for fluorescent detection of Fast Red as advised by ACD. Identical acquisition settings were applied for all images. Z-stacks were orthogonally projected and then quantified using Fiji. Z18 zinc-finger domain structural modeling The protein sequence for Z18 was downloaded (UniProt: Q86VM9) and the zinc-finger domain spanning residues 219 to 245 was selected for homology modeling. The Z18 zinc finger three-dimensional structure was modeled using the online Swiss-Model server82 with the RBM22 protein from the human spliceosome cryo-EM structure with RNA oligomer (PDB: 6ID145) as a template. Modified PAR-CLIP The aromatic zinc finger (ZnF) mutations (F226A, F227A, and W234A) were introduced into Z18trunc using in vitro mutagenesis (NEB, E0554S). A prior study of the CCCH domain in the structurally similar ZC3H12 protein identified the aromatic residue equivalent to F240 to be unimportant for RNA binding; thus, we left F240 intact.89 We were uncertain whether F226 or F227 was important because they are both conserved, so we mutated both. V5-tagged Clover (control), Z18fl, Z18trunc, and Z18trunc ZnF aromatic mutant (hereafter Z18truncZnFmut) constructs were transiently transfected into A375 melanoma cells (Thermo Fisher, L3000008). PAR-CLIP was performed as published46 except that protein size selection was not completed. At 24 h post-transfection, cells were labeled with 100μM 4sU for 16 h, media was removed, and cells were covered with PBS for UV crosslinking (2 min, 316 nm, Spectrolinker XL-1500). The crosslinked cell pellet was lysed with RIPA buffer (50 mg cell pellet/mL RIPA buffer), vortexed, and incubated on ice for 20 min before brief sonication (Covaris E220 ultrasonicator, 1 min, PIP 140.0, Duty factor 7.0, CPB 300). Protein G Dynabeads (100 μL/IP, Invitrogen, 10004D) were conjugated to anti-V5 (20 μg/IP, Sigma, V8012). IP was done in triplicate overnight at 4°C. Beads were washed once with RIPA buffer and twice with IP wash buffer (150 mM NaCl, 50 mM Tris). On-bead Proteinase K digestion (10 μL/IP, NEB, P8107S) was performed at 50 C for 45 min with gentle mixing every 5 min. RNA was isolated via acid phenol-chloroform extraction, purified using NEBNext RNA sample purification beads (NEB, E6351S), rRNA-depleted (NEB, E7405), and prepared for stranded Illumina sequencing (NEB, E7760S). PAR-CLIP data analysis Reads were trimmed and quantified using SQuIRE28 as above. TEs that were significantly bound to bait protein (Z18fl, Z18trunc, or Z18truncZnFmut) versus control Clover were identified using DESeq2.33 Identification of bound regions. Trimmed reads were mapped to the human genome build hg38 using Bowtie (v 1.1.2) in paired-end mode with parameters: -v 2 -m 10 –best –strata.72 To identify high-confidence clusters of T-to-C transitions resulting from UV crosslinking of 4sU-labeled RNA to Z18, wavClusteR version 2.36.048 was used. To identify a unified set of regions of interest for each condition, the wavClusteR output “cluster.bed” files were merged for replicates of each condition (Clover, Z18fl, Z18trunc, or Z18truncZnFmut). Cluster annotation. The combined cluster files for each condition were annotated using the bedtools intersect -c option, for: (a) LTRs and LINEs using UCSC RepeatMasker (downloaded 11/08/2023), (b) exons from refGene.gtf (downloaded 11/08/2023), and (c) introns. Introns were annotated using a.bed file including all disjoint segments of canonical introns90,91 using hg38.ncbiRefSeq.gtf92 (entries referring to patches or haplotypes were removed). The number of clusters annotated as overlapping with LTRs, LINEs, introns, or exons was calculated and plotted for each condition. The average cluster size (base pairs) for each condition was similar (Z18fl 137.51; Z18trunc 132.06; Clover 129.55). Quantification of bound regions. Trimmed reads were remapped to the human genome build hg38 using the splice-aware mapper STAR (v. 2.5.3a) in paired-end mode with parameters –outSAMtype BAM SortedByCoordinate –outWigStrand Stranded –outWigType bedGraph.80 To identify reads that overlapped Z18fl-bound regions from wavClusteR, each BAM was intersected with the merged Z18fl using bedtools -intersect. StringTie v. 1.3.3b81 was used to count expression from intersected BAM files annotated for: (a) LTRs and LINEs, (b) exons, (c) introns, (d) PROMPTs, or (e) eRNAs. Normalized reads per transcript were used as input for DESeq233 using a size factor of 1, comparing Z18fl or Z18trunc versus Clover (control). The log2 fold change of Z18fl or Z18trunc versus Clover for LTRs, LINEs, introns, or exons was plotted, and a nonparametric two-sided Wilcoxon rank-sum test was used to compare clusters annotated to contain LTRs, LINEs, or introns versus exons. Motif analysis. Simple enrichment analysis (SEA)49 was performed using the web server at https://meme-suite.org on merged wavClusteR48 PAR-CLIP Z18fl-bound regions (n = 3) with merged Clover (n = 3) as background and the human HOCOMOCO v11 motif database.93 Motifs identified by SEA were filtered to retain those with the highest true-positive rate and log10 q-value <−10. The top 12 motifs were selected, and nucleotide composition was calculated. Nucleotide base composition analysis. The genomic coordinates for each cluster.bed file were used to calculate the coverage in the corresponding BAM files using mpileup from SAMtools.79 Then the cluster nucleotide composition was calculated. CRISPRi We used the CRISPRi pXPR_066 all-in-one plasmid, which expresses dCas9-KRAB and gRNA sequences in the same vector, obtained from the Genetic Perturbation Platform (GPP, Broad Institute).69 gRNAs were designed to target the most highly regulated ERV RNA, chr14 MER4-int using CRISPOR,74 and a control gRNA targeting an intergenic site used by GPP as a control. Control gRNA and MER4-int gRNA1 or gRNA2 were cloned into the above vector, and gRNA efficiency was assessed via bulk transient transfection (Lipofectamine 3000, Invitrogen, 100022050). Z18trunc-expressing A375 cells were transfected, grown for 36 h, then selected with puromycin for 48 h. RNA was isolated, and qPCR for MER4-int was performed. After confirming gRNA efficacy, the three gRNA constructs were used to generate lentivirus, which was transduced into independently derived Clover- and Z18trunc-expressing A375s (n = 3 each). After 48 h, cells were selected with puromycin for 72 h, grown for 4 days without antibiotics, washed with 1x PBS, fixed with 10% formaldehyde for 10 min, stained with crystal violet for 30 min at RT, rinsed with water, and dried overnight. Images were acquired using a Nikon SMZ18 stereo microscope. Crystal violet was solubilized in methanol, and optical density was measured at 570 nm with a CLARIOstar plate reader. Primers in Table S4. QUANTIFICATION AND STATISTICAL ANALYSIS The statistical details for all analyses are included in either the methods or in the relevant figure legends, including the statistical test used, the value of n, what n represents, and data dispersion measures. Supplementary Material SUPPLEMENTAL INFORMATION Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2026.117526.