Cancer cells subvert the primate-specific KRAB zinc finger protein ZNF93 to control APOBEC3B

Significance

Cells must balance genome stability to prevent harmful genotoxic stress while allowing adaptation. Our study identifies ZNF93, a primate-specific transcription factor, as a key regulator of APOBEC3B, a mutagenic enzyme linked to cancer progression. While APOBEC3B promotes tumor evolution, excessive activity causes DNA damage. We show that ZNF93 mitigates this stress, helping cancer cells survive. Tumors appear to exploit this mechanism, fine-tuning ZNF93 expression to regulate APOBEC3B. Our findings uncover a layer of genome regulation at the intersection of evolution, epigenetics, and cancer biology. Understanding how cancer cells manipulate this balance provides insights into tumor adaptation and potential therapeutic targets.

Abstract

ZNF93 is a primate-restricted Krüppel-associated box zinc finger protein responsible for repressing 20- to 12-My-old L1 transposable elements. Here, we reveal that ZNF93 also regulates the key cancer driver APOBEC3B—a mutagenic enzyme linked to tumorigenesis and cancer progression. ZNF93 depletion impairs DNA synthesis, activates replication and DNA damage checkpoints, and triggers proinflammatory phenotypes. Conversely, its overexpression enhances resistance to exogenous genotoxic stress, mirroring the effects observed with APOBEC3B depletion. ZNF93 expression correlates with cell proliferation rates and is overexpressed in many cancer types. These findings suggest that ZNF93 serves as a critical guardian of genome integrity, co-opted by cancer cells to counterbalance APOBEC3B-induced and L1-derived genomic instability and inflammation.

The human genome harbors >5 million inserts derived from transposable elements (TEs), which together contribute at least 60% of its DNA content (1). Among them are more than 1.5 million Long Interspersed Nuclear Element (LINE) integrants belonging in their majority to the LINE-1 (L1) subfamily (2). L1s are autonomous retrotransposons encoding the machinery necessary for their transposition, a “copy-and-paste” process whereby an RNA intermediate is reverse transcribed into a DNA copy that integrates back into the genome (3). Integration relies on the endonuclease activity of the L1-encoded ORF2p protein, contained in its N-terminal region, inducing DNA double-strand breaks as a first step (4). The C-terminal part of ORF2p then mediates reverse transcription, using the freed 3′ end of a DNA strand as a primer and the L1 RNA as a template (3).

L1s predate the origin of vertebrates and in mammals display a ladder-like phylogeny where only the youngest group is active at any given time, expanding from a common progenitor for a few million years before being replaced by a new group (2, 5). This pattern is consistent with an arms race between a host that evolves functions repressing L1 transposition, and L1 that mutates in return to bypass this obstacle (6). In the human lineage, L1 restriction factors are exemplified by ZNF93, a member of the Krüppel-associated box (KRAB)-containing zinc finger proteins (KZFP) family, the largest group of DNA-binding proteins in higher vertebrates (7, 8). There are close to 400 human KZFPs, the great majority of which recognize a particular TE subgroup as primary genomic target via a C-terminal array of zinc fingers, allowing for sequence-specific DNA binding (7–10). 80% of human KZFPs act as transcriptional repressors through recruitment by their N-terminal KRAB domain of a KAP1/TRIM28-nucleated heterochromatin-inducing complex while the remaining, often older family members display other functionalities linked to the presence of variant KRAB or additional domains (7, 8, 10, 11). ZNF93 is a canonical KZFP that emerged some 20 Mya, contemporary to the L1PA6 L1 subfamily, which it represses by recognizing a sequence located close to the 5′ end of their integrants (6). The ZNF93 target motif is conserved in the more recent L1PA5 and L1PA4 integrants but was lost through a 129 bp-long deletion that occurred in some of their L1PA3 descendants, allowing these and their L1PA2 and L1Hs progeny to escape ZNF93-mediated control (6). While the ZNF93-L1 interplay is a convincing illustration of the arms race model, it poses a conundrum. Indeed, only ~100 human L1 integrants are still replication-proficient, all belonging to the less than 3 My old L1Hs subgroup (12). The TE targets of ZNF93, like those of an overwhelming majority of KZFPs, are thus transposition-incompetent, due to the accumulation of inactivating mutations over time, yet they remain bound by their cognate KZFP (6).

Cancer represents an interesting setting to study the interplay between TEs and their epigenetic controllers. In cancer cells, epigenetic alterations typically trigger the derepression of TEs, which leads to type I interferon response (13, 14). Cell clones are then selected that express high levels of epigenetic repressors, which reinstate TE silencing, thereby mitigating replication stress and allowing cells to resume growth (15, 16). L1s have been identified as proinflammatory in the context of both cancer and cellular senescence (17, 18), whether due to loss of epigenetic repression in the former (16, 18), or to the effect of transcriptional activators in the latter (17). In both situations, increases in L1 expression correlate with elevated DNA damage and replication stress (16–18), part of which may be due to the endonuclease activity of ORF2p (4, 19, 20).

Here, we examined how L1-targeting KZFPs might influence this process. This led us to identify ZNF93 as a family member almost systematically upregulated in cancer cells. We further determined that many of the L1 targets of this KZFP encode a truncated form of ORF2p with increased endonuclease activity. Accordingly, we observed that ZNF93 depletion led to genomic instability and inflammation in cancer cells. However, we unexpectedly found that much of this effect stemmed from the loss of ZNF93-mediated repression of APOBEC3B, a cytidine deaminase known not only as a restriction factor of L1 (21–24) and various viruses (25) but also as an inducer of cell death upon DNA damage and replication stress (26–33) and a well-documented mutagen in cancer (30, 31, 34–38). These results position ZNF93 as a critical guardian of genome stability subverted by cancer cells to foster tumor growth.

Result

KZFPs Accumulate at the Promoter of Endonuclease-Proficient Young LINE-1 Integrants.

To identify KZFPs responsible for preventing L1 ORF2p-induced genotoxicity, we first conducted a census of KZFP binding over endonuclease-encoding (EN+) L1 elements. Most L1-derived sequences are 5′-truncated, thereby lacking the 5′ UTR-embedded promoter hence the ability to express L1-encoded proteins. We thus mapped KZFP binding over EN+ full-length L1s, defined as L1 integrants spanning at least 5.9 kb and harboring an ORF2p open reading frame longer than 230 amino acids, the minimal length of its endonuclease moiety. We focused on the seven youngest L1 (YL1) subgroups, L1PA7 to L1HS (2), which together accounted for most of the full-length L1s (6,691/8,597, 78%). We further determined that due to the accumulation of inactivating mutations, only 28% (1,842/6,691) of full-length YL1 (FLYL1) are EN+ (Fig. 1A).

Fig. 1.

KZFPs accumulate at the promoter of endonuclease-proficient young LINE-1 integrants. (A) Full-length (>5,900 bp) and recently evolved (L1PA7 to L1HS) LINE-1 elements (FLYL1s), with (EN+) or without (EN−) putative endonuclease activity. (B) KRAB-KZFP binding enrichment (Fischer’s exact test, adj. P < 0.05) at FLYL1 elements endowed with endonuclease activity (FLYL1^EN+). The KRAB score represents the probability of recruiting TRIM28 (39). Data from refs. 7, 9, 10. (C) ChIP signal for the KZFPs identified in (B) across the meta-FLYL1^EN+ sequence.

Using previously compiled genome-wide binding profiles (7, 9, 10) we identified 15 KZFPs, the binding of which was enriched at FLYL1^EN+ loci (Fig. 1B and SI Appendix, Fig. S1A). All bore KRAB domains predicted to induce TRIM28-dependent H3K9me3 deposition (39), consistent with their potential role as FLYL1^EN+ repressors. KZFP binding was particularly dense at FLYL1^EN+ promoters, with ZNF649, ZNF93, ZNF765, ZNF577, and ZNF141 exhibiting the highest enrichments (Fig. 1C). Interestingly, KZFPs displayed preferential binding to specific regions of the FLYL1^EN+ promoter. For example, ZNF93 predominantly targeted the 3′ end of the promoter, while ZNF649 localized significantly upstream in the 5′UTR and ZNF765 somewhere between them. Some KZFPs recognize regions outside of the promoter, such as ZNF382 and ZNF84, which bind to distant sequences in ORF2 (Fig. 1C).

Moreover, the KZFP-mediated recognition of FLYL1^EN+ displayed remarkable subfamily specificity (SI Appendix, Fig. S1A). For instance, ZNF765 and ZNF93 showed maximal enrichment at L1PA5^EN+ and L1PA4^EN+ promoters, respectively, while ZNF382 and ZNF84 bound most FLYL1^EN+ including from the L1HS subset. Also, L1PA7^EN+ and L1HS^EN+—representing the oldest and youngest subfamilies of FLYL1, respectively—were recognized by the narrowest array of KZFPs (SI Appendix, Fig. S1A). This pattern likely reflects on the one hand the progressive degeneration of KZFP-binding motifs in older subfamilies like L1PA7, and on the other hand, insufficient evolutionary time for the selection and fixation of L1HS-specific KZFP repressors. Together, these results pointed to ZNF93, ZNF649, and ZNF765 as strong candidate FLYL1^EN+ repressors.

ZNF93 Is Expressed in Proliferating Cells, Marking TRIM28 Recruitment at FLYL1^EN+ Promoters.

Increased expression of epigenetic repressors is frequently selected for in cancer cells, resulting in reduced TE expression, prevention of DNA damage, replication stress, and inflammation, all of which partake in sustaining proliferation (15, 16, 18). Consequently, we reasoned that among FLYL1^EN+-enriched KZFPs, those expressed concomitantly with proliferation would likely be the most functionally significant FLYL1^EN+ repressors for cancer evolution.

To test this hypothesis, we evaluated the correlation between a 167-gene proliferation signature and the expression of FLYL1^EN+-enriched KZFPs across 23 cancer subtypes from The Cancer Genome Atlas (TCGA) (39). ZNF93 displayed the most pronounced (Wilcoxon two-tailed test, P < 0.05) and frequent (14/23 cancer subtypes) statistically significant (BH-adj. P < 0.05, t test) association with proliferation (Fig. 2A). Notably, ZNF93 was only topped by ZNF695 and ZNF724P when considering expression-versus-proliferation correlations across all KZFPs and TCGA tumor samples (SI Appendix, Fig. S2A). Moreover, ZNF93 displayed a greater tendency for upregulation in cancer versus normal tissue compared to other FLYL1^EN+ promoter-enriched KZFPs (Fig. 2A). Together, these results show that ZNF93 expression is correlated with proliferation in cancer.

Fig. 2.

ZNF93 is expressed in proliferating cells, marking TRIM28 recruitment at FLYL1^EN+ promoters. (A, Top) Spearman rank correlation (t test, adj. P < 0.05) between KZFP expression and a 167-gene proliferation signature across 33 TCGA cancer subtypes. Red: significant correlation; Blue: significant anticorrelation. (Bottom) Differential KZFP expression between cancer and normal samples (23 cancer subtypes selected). Median Spearman correlation coefficients and log2FC were compared across subtypes using the two-tailed t test. (B) KZFP expression in resting, IL-7 activated, and TCR-activated CD4⁺T cells. Means across two independent experiments are shown. Differential expression analysis was performed. (C, Left) ChIP signal for KZFPs across the meta-FLYL1^EN+ promoter sequence. (Right) TRIM28 ChIP signal in resting, IL-7-activated and TCR-activated CD4⁺T cells.

To explore further the link between FLYL1^EN+-enriched KZFPs and proliferation, we analyzed the differential responsiveness of resting primary CD4⁺T cells to IL-7, a nonmitogenic T cell activator, versus CD3/CD28 crosslinking, a strongly mitogenic T cell activator [public dataset (40), SI Appendix, Fig. S2B]. Genes upregulated by TCR crosslinking, but not IL-7, were enriched for cell cycle and proliferation markers (SI Appendix, Fig. S2 C and D). While IL-7 had little to no effect on the expression of FLYL1^EN+-enriched KZFPs, most (13/15) were either unaffected or downregulated upon TCR crosslinking. In contrast, ZNF93 was robustly upregulated upon TCR crosslinking (Fig. 2B). This result reinforces the association between ZNF93 expression and proliferation.

Next, we investigated whether binding of the KZFP cofactor TRIM28 to FLYL1^EN+ loci was altered upon mitogenic versus nonmitogenic T cell activation. Strikingly, TRIM28 occupancy specifically increased at ZNF93 binding sites upon TCR crosslinking but not IL-7 treatment. In contrast, IL-7 treatment increased TRIM28 binding at ZNF649 and ZNF765 binding sites (Fig. 2C and SI Appendix, Fig. S2E). These data indicate that among FLYL1EN+-enriched KZFPs, ZNF93 most specifically correlates with TRIM28 recruitment at L1 promoter under mitotic stimulation.

ZNF93 Depletion Reduces Proliferation and Induces Replicative Stress in Cancer Cells.

To assess the role of ZNF93 in proliferating cells, we depleted this KZFP through RNA interference, achieving a fourfold knockdown (KD) (Fig. 3A) in acute myeloid leukemia-derived K562 cells. ZNF93 KD reduced the proliferation rate of K562 by ~40% (Fig. 3B). This phenotype was consistently reproduced across multiple cell lines, including another leukemia cell line (THP1), three B cell lymphoma cell lines (OCY19, SUDHL4, and U2932) and two colorectal cancer cell lines (LS1034 and SW480), each expressing various levels of the endogenous ZNF93 at baseline (Fig. 3 C and D and SI Appendix, Fig. S3 A and B).

Fig. 3.

ZNF93 depletion reduces proliferation in cancer cells. (A) ZNF93 expression levels (RT-qPCR) upon ZNF93 KD in K562 cells or scrambled (scr) (n = 6, t test). (B) Metabolic activity as a surrogate for proliferation of K562 cells upon ZNF93 KD (n = 6, t test). (C) ZNF93 expression in cancer cell lines from (https://sites.broadinstitute.org/ccle/). Cell lines used in this study are highlighted in red for cell lines expressing ZNF93 and blue for cell lines not expressing ZNF93. (D) ZNF93 KD efficiency and associated effects on cell proliferation across cell lines. (E) Flow cytometry analysis of the replication signal (EdU incorporation intensity, y-axis) and DNA content (DAPI staining, x-axis) upon ZNF93 KD in K562 cells. (F) pATR, pATM, pCHEK2, pCHEK1, and γH2AX signals upon ZNF93 KD in K562 cells. (G) Heatmap plot of Normalized Enrichment Score (NES) of Hallmark signatures derived from RNA-seq data upon ZNF93 KD in the OCI-LY7, SW480, K562, and U2932. Asterisks indicate significant enrichment compared to control conditions.

To ascertain that the observed antiproliferative effects were specific to ZNF93 depletion and not due to off-target activity of the short hairpin RNA (shRNA), we used HeLa and HCT116 cell lines, which naturally lack detectable ZNF93 expression (Fig. 3C). Compared to SW480—the cell line with the lowest detectable ZNF93 among those exhibiting a proliferation phenotype—ZNF93 transcript levels were found to be >50-fold lower in HCT116 and >200-fold lower in HeLa, as measured by RT-qPCR. As anticipated from such a setting, no further decrease in ZNF93 levels could be detected upon introduction or the ZNF93-targeting shRNA (SI Appendix, Fig. S3D). Most importantly, this manipulation also did not slow down the proliferation of these cells (Fig. 3D and SI Appendix, Fig. S3E), very strongly suggesting that the growth impairment observed when ZNF93 was depleted from cells expressing this factor was not due to off-target effects.

Together, these results show that ZNF93 depletion induces replication stress, activates the DNA damage response and inflammation, which result in stalling proliferation in multiple cancer cell lines.

ZNF93 Depletion Modestly Derepresses FLYL1 and Fails to Induce Detectable Changes in L1-Encoded Protein Expression.

We observed that ectopic overexpression (OE) of ORF2p, but not a catalytically dead endonuclease mutant, recapitulated the phenotypes observed upon ZNF93 depletion. Confirming previous results (4), ORF2p OE led to reduced K562 proliferation (SI Appendix, Fig. S4 A and B), impaired EdU incorporation in S-phase (SI Appendix, Fig. S4C), and pCHK1/2 accumulation (SI Appendix, Fig. S4D). Given the similarity between the DNA damage response and replication stress phenotypes induced by ZNF93 KD and ORF2p OE, we investigated whether the former resulted from an accumulation of ORF2p upon FLYL1^EN+ derepression. To this end, we characterized the epigenomic and transcriptional changes occurring at FLYL1^EN+ loci upon ZNF93 KD. Cleavage Under Targets and Tagmentation (CUT&Tag) experiment revealed only modest reductions in H3K9me3–the repressive heterochromatin mark instigated by the KZFP/TRIM28 axis—at ZNF93-bound FLYL1^EN+ promoters, with many FLYL1^EN+ loci remaining unaffected (Fig. 4A). RNA-seq analyses revealed that ZNF93 KD induced FLYL1 upregulation to a degree that varied between different cell lines (Fig. 4B and SI Appendix, Fig. S4 E and F). However, this was not limited to ZNF93 L1 targets, as L1Hs and L1PA2 integrants also exhibited increased transcription (SI Appendix, Fig. S4G).

Fig. 4.

ZNF93 depletion modestly derepresses FLYL1 but fails to induce detectable changes in L1-encoded protein expression. (A, Left) ZNF93-HA ChIPexo signal upon OE in HEK293T cells (9) and H3K9me3 ChIP-seq signal upon ZNF93 KD in K562 cells across FLYL1^EN+ promoters. (Right) H3K9me3 signal across the meta-FLYL1^EN+ sequence upon ZNF93 KD in K562 cells. (B) Normalized RNA fold change across TEs upon ZNF93 KD in K562, OCI-LY7, SW480, and U2932. Statistical test: two-tailed t test. (C) ORF1p signal upon ZNF93 KD in K562 cells. Quantification relative to Ponceau S Staining on the right.

This modest transcriptional activation of ZNF93-targeted FLYL1 elements, despite the loss of ZNF93, suggests that additional KZFPs, particularly ZNF765 and ZNF649 may contribute to their repression in a redundant manner. To test this hypothesis, we conducted TRIM28 Chromatin Immunoprecipitation (ChIP)-qPCR at A L1PA4 upon ZNF93 KD. While lost at the ZNF93 binding site, TRIM28 is maintained and even tends to increase at ZNF649 and ZNF765 binding sites upon ZNF93 KD, pointing to a redundant and possibly compensatory in-between KZFPs (SI Appendix, Fig. S4H). This redundancy may help explain why ZNF93-targeted L1 elements do not exhibit robust transcriptional activation in its absence. Together, these findings highlight the cooperative nature of KZFP-mediated silencing at FLYL1 loci.

We sought to determine whether FLYL1 upregulation resulted in increased intracellular levels of the genotoxic, L1-encoded ORF2p. We successfully detected L1-encoded endogenous ORF1p by western blot, but found its expression to be unaffected by ZNF93 KD (Fig. 4C). In spite of multiple attempts, we failed to detect endogenous ORF2p, whether in control or in ZNF93 KD cells, consistent with the minute levels at which ORF2p is expressed relative to ORF1p (41, 42). Thus, we could not firmly establish that the observed DNA damage and replication stress observed upon ZNF93 depletion resulted from the uncontrolled expression of L1 ORF2p.

ZNF93 Is a Critical Regulator of APOBEC3B Expression.

Although ZNF93 primarily binds to L1 5′UTRs, it is also modestly enriched at some gene promoters, suggesting a potential role in regulating gene expression via promoter occupancy (SI Appendix, Fig. S5A). We thus examined transcriptional changes occurring upon ZNF93 KD by combining RNA-seq data from ZNF93 KD and ZNF93 OE experiments in K562 cells (SI Appendix, Fig. S5B), focusing on changes affecting genes with a ZNF93 ChIPseq peak within 50 kb window of their transcription start sites (TSS). This analysis identified the gene encoding APOBEC3B—a cytidine deaminase known to act as a DNA editor and inducer of replication stress and DNA damage—as a top ZNF93-regulated target. APOBEC3B was significantly upregulated upon ZNF93 KD and downregulated upon ZNF93 OE, and further displayed a prominent ZNF93 binding peak at its TSS (Fig. 5A).

Fig. 5.

ZNF93 is a critical regulator of APOBEC3B expression. (A) Gene expression changes upon ZNF93 KD (x-axis) and OE (y-axis) in K562 cells. Orange dots indicate ZNF93 peaks within 50 Kb of the TSS, while black dots indicate ZNF93 peaks overlapping the TSS. (Left) Fold changes relative to control. (Right) -Log10 P-value. (B) APOBEC3B expression levels after ZNF93 KD in K562, U2932, SW480, and OCI-LY7 cells, as well as 5 d after ZNF93 OE induced by doxycycline in K562 cells. Statistical test: Differential expression. (C) APOBEC3B signal 6 d after ZNF93 KD in K562 cells. (D) APOBEC3B signal 5 d after OE of GFP-HA or ZNF93-HA in K562 cells. (E) Integrated Genome Browser (IGB) screenshots showing the ZNF93 ChIP-exo track in 293 T cells (9), TRIM28 ChIP-seq in H1 cells, and H3K9me3 and H3K4me3 CUT&TAG signals after ZNF93 KD in K562 cells (sum of four independent biological replicates), at the APOBEC3B TSS and at one representative L1PA4. (F) ChIP–qPCR analysis of TRIM28 binding at the APOBEC3B TSS after OE of GFP or ZNF93 in K562 cells. Means from two independent experiments are shown. Neg and Pos are negative and positive control regions for TRIM28 binding. (G, Top) mutagenesis of ZNF93 (ZNF93^Mut) with four amino acid substitutions in the KRAB domain, abrogating TRIM28 recruitment. (Bottom) HA signal after OE of HA-tagged ZNF93, ZNF93^Mut, or GFP in K562 cells. (H) Coimmunoprecipitation of TRIM28 with HA-tagged proteins as described in Fig. 5G. HA was immunoprecipitated from K562 cell lysates 6 d after OE of ZNF93, ZNF93^Mut, or control (GFP). (I) ChIP–qPCR analysis of TRIM28 binding at the APOBEC3B TSS after OE of HA-tagged ZNF93, ZNF93^Mut, or GFP in K562 cells. Means from two independent experiments are shown. (J) RT-qPCR analysis of APOBEC3B expression 5 d after OE of HA-tagged ZNF93, ZNF93^Mut, or GFP in K562 cells. Means and SD from three independent experiments are shown. *P < 0.05 (two-tailed t test). (K) APOBEC3B signal after OE of HA-tagged ZNF93, ZNF93^Mut, or GFP in K562 cells.

Integrating RNA-seq data from additional cell lines (U2932, OCY7, and SW480) (SI Appendix, Fig. S5C) revealed that APOBEC3B was in fact the only gene behaving as a strict ZNF93 target, being consistently upregulated across all ZNF93 KD conditions (Fig. 5B and SI Appendix, Fig. S5C). Accordingly, ZNF93 KD led to a dramatic increase in APOBEC3B protein levels, which were conversely reduced in ZNF93 OE K562 cells (Fig. 5 C and D).

We next assessed whether the ZNF93-APOBEC3B regulatory dependency correlated with epigenetic changes at the APOBEC3B promoter. There was surprisingly no clear H3K9me3 peak at the APOBEC3B ZNF93 binding site (Fig. 5E), where TRIM28 was not recruited, in sharp contrast with what observed at FLYL1 promoters (Fig. 5 E and F). However, there was an increase in the activation mark H3K4me3 at the APOBEC3B promoter in ZNF93 KD K562 cells (Fig. 5E). This strongly suggested that the KZFP regulated the cytidine deaminase by nonconventional mechanisms.

To test this hypothesis, we mutated ZNF93 at residues of its KRAB domain previously determined to be essential for TRIM28 recruitment (Fig. 5G) (43). Coimmunoprecipitation (Co-IP) confirmed that the resulting ZNF93^mut indeed failed to associate with the master corepressor (Fig. 5 H and I). Nevertheless, upon OE, this mutated protein still silenced APOBEC3B expression, as verified at both RNA and protein levels, albeit slightly less efficiently than its wild-type counterpart (Fig. 5 J and K). Therefore, ZNF93 represses APOBEC3B expression largely through TRIM28-independent mechanisms.

ZNF93 Protects Proliferating Cells from APOBEC3B-Induced Replication Stress.

APOBEC3B expression is known to increase in response to cancer cell proliferation and genotoxic stress, and in return to amplify replication stress (26–33). Having established that ZNF93 silencing upregulates APOBEC3B, we hypothesized that ZNF93 may limit replication stress in proliferating cells by repressing the cytidine deaminase. To test this hypothesis, we characterized the impact of ZNF93 on cellular responses to the genotoxic stressor hydroxyurea (HU). As previously observed (31), HU treatment led to APOBEC3B induction and increased γH2AX levels, indicative of DNA damage and replication stress. This response was markedly reduced by ZNF93 OE (Fig. 6A). Furthermore, while ZNF93 OE could not alleviate the HU-induced replication block (SI Appendix, Fig. S6A), it allowed for a swifter return to proliferation following HU washout, similar to what observed upon APOBEC3B KD (Fig. 6B and SI Appendix, Fig. S6B), whereas the opposite was observed upon APOBEC3B OE (Fig. 6B and SI Appendix, Fig. S6C).

Fig. 6.

ZNF93 protects proliferating cells from APOBEC3B-induced replication stress. (A) APOBEC3B and γH2AX signal 3 d after induction of GFP-HA or ZNF93-HA, followed by HU treatment for 20 h in K562 cells. (B, Left) EdU-fluorescence-activated cell sorting (FACS) analysis 3 d after induction of GFP-HA, APOBEC3B-HA, or ZNF93-HA, or after shRNA-mediated depletion of APOBEC3B, followed by HU treatment for 20 h and an 8-h recovery period in K562 cells. (Right) Quantification of EdU+ cells. Statistical significance: **P < 0.01; ***P < 0.001 (two-tailed t test).

Discussion

Our study identifies the primate-specific KZFP ZNF93 as a pivotal guardian of genome stability. While ZNF93 has been recognized for its role in silencing TEs, particularly L1 retrotransposons (6), we provide evidence that it more broadly mitigates replication stress in proliferative or genotoxic contexts by also repressing the expression of the APOBEC3B cytidine deaminase.

APOBEC3B, a primate-restricted member of the APOBEC family of viral restriction enzymes, uniquely locates to the nucleus, where it restricts both retrotransposons and viruses through DNA editing. In addition, APOBEC3B has emerged as a critical regulator of secondary structures known as R-loops during DNA replication (44). Fittingly, APOBEC3B exhibits a robust cell cycle-rhythmic expression pattern, reaching maximal expression during the S phase (39). Thus, APOBEC3B, similar to SAMHD1 and other innate immune-related proteins, stands at the crossroads of immunity and DNA replication (45–47), bridging the oft-mentioned divide separating fast-evolving, environment-interacting genes—e.g., immune genes—from ultraconserved genes involved in ubiquitous processes such as the cell cycle (48). Related to this, we have recently reported that surprisingly many KZFPs—given their recent evolutionary emergence—regulate cell cycle-oscillating genes via promoter binding, suggesting unappreciated levels of lineage specificity in the regulation of ubiquitous cellular processes (39). The existence of a wholly primate-specific ZNF93-APOBEC3B regulatory axis further indicates that fundamental biological processes are not solely carried out by deeply conserved genes.

APOBEC3B, together with APOBEC3A, has been identified as a potent host mutagen in multiple cancer cell types, contributing to tumor heterogeneity, evolution, and resistance to therapy (30, 31, 37, 49, 50). Yet, unabated levels of APOBEC3B-mediated mutagenesis may result in unsustainable, fatal levels of replication stress and DNA damage (51). The high correlation observed between ZNF93 expression and cell proliferation across most cancer types suggests that by repressing APOBEC3B, ZNF93 partakes in this trade-off. The ZNF93-APOBEC3B regulatory dependency may therefore have important implications for tumor progression. Supporting this, the only cancer type where ZNF93 expression anticorrelates with proliferation according to TCGA (BH-adj. P < 0.05, rho < 0) is cervical squamous cell carcinoma (CESC). CESC is frequently caused by human papillomavirus (HPV) infection, which APOBEC3B participates in restricting. This antiviral action results in the collateral accumulation of mutations, thereby favoring cancer initiation. Accordingly, we noted that compared to HPV-negative tumors, HPV-positive tumors, which express high levels of APOBEC3B (52), simultaneously exhibit low ZNF93 expression (Fig. 7). It is thus tempting to speculate that in cancer cells, regulating ZNF93 expression may be adaptative and partake in balancing APOBEC3B-mediated mutagenesis against APOBEC3B-induced replication stress. Interestingly, ZNF93 has been found to confer human chondrosarcoma cells with resistance to the inducer of replication stress and DNA damage ET-743 (Trabectedin; Yondelis®) (53), an effect potentially linked to ZNF93-mediated repression of APOBEC3B.

Fig. 7.

ZNF93 is downregulated in HPV-positive cervical cancer samples. (A, Left) differential KZFP expression between cancer and normal samples (23 cancer subtypes selected). Median Spearman correlation coefficients and log2FC were compared across subtypes using Wilcoxon’s two-tailed test. (Right) Spearman rank correlation (t test, adj. P < 0.05) between KZFP expression and a 167-gene proliferation signature across 33 TCGA cancer subtypes. Red: significant correlation; Blue: significant anticorrelation. CESC position is annotated. (B) Comparison of ZNF93 and APOBEC3B mRNA expression across HPV-positive (HPV+), HPV-negative (HPV−), in the TCGA cervical (CESC) cohorts. Expression data (Level 3) was obtained from the Broad Genome Data Analysis Center’s Firehose server.

ZNF93-mediated regulation of APOBEC3B appears partially independent of TRIM28, despite robust TRIM28 recruitment at ZNF93-bound TEs. Similar dissociations between TRIM28 and KZFP-mediated promoter regulation have been observed previously (54), suggesting that KZFPs may employ distinct mechanisms at gene promoters compared to TEs. Future work should explore whether steric hindrance, for example competition with the binding of specific transcription factors, underlies this TRIM28-independent regulatory activity, and the mechanisms responsible for the failure of canonical KZFPs to recruit the corepressor at some but not other genomic loci.

Questions remain regarding the impact of ZNF93-mediated control of FLYL1. Upon observing the genotoxic effect of ZNF93 depletion, we first thought that it resulted from the production of ORF2p endonuclease by ZNF93-repressed FLYL1 integrants. We speculated that the preservation, in some cases for more than 20 My, of the toxic endonuclease-coding potential of thousands of integrants dispersed throughout the genome allowed the cell to use them as sentinels watching for loss of epigenetic control. However, we only observed a mild upregulation of FLYl1s in the ZNF93 KD cancer cell lines tested here, with minimal changes in H3K9me3 on their 5′UTR, no discernable elevation in ORF1p levels, and no detectable ORF2p. While this could be attributed to the effect in these cells of other KZFPs binding to the promoter of these L1 integrants, such as ZNF649 and ZNF765, we cannot formally exclude that minute amounts of highly active endonuclease, which is physiologically produced at levels between 200 and 500 lower than ORF1p, were partly responsible for part of the observed phenotype, nor that this protein plays a more prominent surveillance role in other settings.

Irrespectively, our data support a model whereby KZFPs and their genomic targets act as rheostats of genomic instability in cancer cells, with their combined actions likely exerting broad influences on DNA damage, replicative stress, DNA editing, and other forms of mutagenesis, inflammation, immune escape, and resistance to some therapies. As recently demonstrated for ZNF417 and ZNF587, the upregulation of which protects diffuse large B cell lymphoma cells by minimizing HERVK-induced genotoxicity, the ZNF93-APOBEC3B regulatory axis represents a primate-restricted determinant of replication stress, underscoring an underappreciated level of lineage-specificity in the management of genome integrity.

Material And Methods

L1 Characterization.

Elements are annotated as full length if the calculated length using hg19 annotation is >5,900 bp. Over full-length L1s, ORF2p were detected using ORF detection (55) with the following parameters: Minimum nucleotide size of ORF to report: 30, Maximum nucleotide size of ORF to report: 10,000, What to output: Translation of regions between START and STOP codons, All START codons to code for Methionine: true, Circular sequence: false, Find ORFs in the reverse complement, true, Number of flanking nucleotides to output: 0, Output sequence file format: FASTA (m).

Among the detected ORFs, L1-ORF2p were detected using the following parameters: starts with “MTGSNSHITI” with at most two-character difference. L1s are annotated as FLYL1^EN+ if >230 aa and FLYL1^EN- if not. Output: SI Appendix, Table S1.

Enrichment of KZFPs on Repeats.

Enrichment for KZFPs over L1s was extracted from ref. 7. Enrichments were computed with pyTEnrich available at URL https://github.com/alexdray86/pyTEnrich. Association with PVal < 0.05 are considered significant and used for further analysis.

Average Profiles of KZFPs on L1 Regions.

Matrixes were generated using deeptools 2 (56): computeMatrix with the L1s coordinates (SI Appendix, Table S1) (-R) and various Bigwigs (-S)”; and the following options -bs 50 –sortRegions keep –sortUsing mean –averageTypeBins mean –outFileSortedRegions”. Then, matrices were used to perform average profiles using plotProfile software from Deeptools2 using the following option: “–averageType mean.”

TCGA Analysis.

Analysis as in ref. 57 with the following tables: SI Appendix, Tables S2 and S3 and Table S3. A proliferation signature consisting of 167 manually curated cell cycle marker genes covering all phases of the cell cycle (57) was used to score tumor samples. Counts per million of signature genes were log-transformed, z-score normalized within each cancer type, and averaged to generate a proliferation score for each sample. Spearman rank correlations were calculated between KZFP expression and proliferation scores within each TCGA cancer type. P-values were adjusted using the Benjamini–Hochberg method, and correlations were considered significant at an adjusted P-value < 0.05.

CD4+T Cell Dataset.

The datasets to analyze RNAseq and KAP1 ChIPseq are public (40) (SI Appendix, Table S4). Gene Ontology enrichment was performed on the DE genes with the following script:

# Load necessary libraries while suppressing warnings and messages

library(clusterProfiler), library(matrixStats), library(gplots), library(RColorBrewer), library(sqldf), library(hopach), library(edgeR), library(limma), library(GOstats), library(GO.db), library(org.Hs.eg.db), library(org.Mm.eg.db), library(data.table), library(circlize), library(gridExtra), library(ggplot2), library(dplyr)}))

# Set new working directory

setwd(“”)

# Load significant genes dataset

Significant_Genes <- read.csv(“Significant_Genes.txt”, sep=””)

# Load normalized expression values

norm_vals <- read.delim(“norm_vals.xls”)

# Merge data based on the “symbol” column

Table_GO <- merge(Significant_Genes, norm_vals, by = “symbol”)

# Save the merged table to a file

write.table(Table_GO, “./Table_GO.txt”, sep=”\t”, row.names = FALSE)

# Load dataset for GO plotting

ForGO_Plotting <- read.delim(“ForGO_Plotting.txt”)

# Load expressed genes (universe data)

exp <- read.csv(“norm_vals.xls”, sep=”\t”, header = TRUE)

universeGeneIds <- exp[exp$symbol != “---”, “symbol”]

# Perform Gene Ontology (GO) enrichment analysis

GOcluster <- compareCluster(X ~ Condition, data = ForGO_Plotting, ont = “BP”, keyType = “SYMBOL”, universe = universeGeneIds,

OrgDb = org.Hs.eg.db, fun =”enrichGO”, pAdjustMethod = “BH”, pvalueCutoff = 0.075, qvalueCutoff = 0.075)

# Define plotting variables

showCategory <- 3 # Number of GO terms to plot per cluster

simplify.cutoff <- 0.7filter.level <- 6

# Standard GO enrichment dot plot

dotplot(GOcluster, showCategory = showCategory)

# Simplify GO terms based on semantic similarity

GOsimpl <- simplify(GOcluster, cutoff = simplify.cutoff, by = “p.adjust”, select_fun = min)

dotplot(GOsimpl, title = “Simplified Semantic Similarity”, showCategory = showCategory)

# Filter GO terms by hierarchical level

GOlevel <- gofilter(GOcluster, level = filter.level)

dotplot(GOlevel, title = “Filtered by Tree Level”, showCategory = showCategory)

# Apply both level filtering and semantic similarity simplification

GOlevel_then_similarity <- simplify(GOlevel, cutoff = simplify.cutoff, by = “p.adjust”, select_fun = min)

dotplot(GOlevel_then_similarity, title = “Filtered and Simplified GO Terms”, showCategory = showCategory)

The list of genes annotated in the proliferation signature (57) was used.

Cell Culture.

OCI-Ly7, OCI-Ly19, and U2932 human lymphoma cell lines were obtained from DSMZ—German Collection of Microorganisms and Cell Cultures, Braunschweig, Germany (www.dsmz.de). SUDHL4 were kindly supplied by Elisa Oricchio (EPFL, Swiss Institute for Experimental Cancer Research) and typed using short tandem repeat profiling by Microsynth cell line authentication service. THP1 were kindly supplied by Caroline Arber [Faculty of Biology and Medicine, University of Lausanne (UNIL). Immuno-Oncology Service (ION)]. K562, LS1034, HCT116, HeLa, and SW480 cell lines were obtained from the American Type Culture Collection. SW480 were cultured in L15 medium (Sigma) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin. K562, THP1, OCI-LY19, U2932, SUDHL4, and LS1034 cells were grown in RPMI 1640 (Gibco) with 10% FBS and 1% penicillin-streptomycin. OCI-Ly7 cells were grown in Iscove’s Modified Dulbecco’s Medium (Gibco) with 20% FBS and 1% penicillin-streptomycin. HeLa and HCT116 were grown in RPMI 1640 (Gibco) with 10% FBS and 1% penicillin-streptomycin. All cell lines were grown at 37 °C and 5% CO2.

Lentivector Production and Stable Gene Expression.

Cells overexpressing HA-tagged constructs were generated as described in ref. 7. In short, complementary DNA (cDNAs) for HA-tagged constructs were codon-optimized, synthesized using the GeneArt service from ThermoFisher, and cloned into the doxycycline (Dox)-inducible expression vector pTRE-3HA. Stable cell lines were generated using lentivector transduction of Mycoplasma-free HEK293T cells as described on http://tronolab.epfl.ch. Sequence ORF ZNF93: Sequence ORF APOBEC3B:

KD Protocol.

Cells were transduced with lentivector particles expressing shRNA targeting the gene of interest (SI Appendix, Table S8). Cells were seeded at a density of 1 × 10⁵ cells/mL and transduced. After 48 h, cells were treated with 1 µg/mL puromycin for 3 d to select for successfully transduced cells. After selection, cells were replated at 1 × 10⁵ cells/mL and collected for further analyses at various time points up to 12 d posttransduction.

RT-qPCR.

Total RNA extraction was performed using the NucleoSpin RNA plus kit (Macherey-Nagel) according to the manufacturer’s recommendations. cDNA synthesis for qPCR was conducted using the Maxima H minus cDNA synthesis master mix (ThermoScientific). Real-time quantitative PCR was performed using PowerUp SYBR Green Master Mix (ThermoScientific) and run on a QuantStudioTM 6 Flex Real-Time PCR System. Primers used are indicated (SI Appendix, Table S8). Enrichment was calculated relative to ALAS1 and TBP expression.

Cell Proliferation Assays for KD Experiments.

Cell proliferation was assessed using the PrestoBlue™ or crystal violet assays. K562, U2932, OCI-LY19, THP1, and SUDHL4 cells were plated 5 d post–lentiviral vector (LV) transduction and after 3 d of puromycin selection. Cells (100 µL) were seeded in duplicate 96-well plates at 20,000 cells/well. PrestoBlue™ (10 µL) was added, and after a 3-h incubation at 37 °C, absorbance was measured at 570 nm, with background subtraction at 600 nm. Absorbance was measured again after 4 d to assess proliferation. For LS1034, SW480, HeLa, and HCT116 cells, plating occurred under the same conditions at 20% confluence in plates and flasks, respectively. For LS1034, SW480, HeLa, and HCT116 cells, proliferation was evaluated using crystal violet staining. Five days post–LV transduction. Cells were seeded in flasks and multiwell plates respectively at 20%, allowed to grow for 3 d, and washed twice with Phosphate Buffer Solution (PBS) after treatment. Fixation was performed with 100% methanol for 10 min at room temperature (RT), followed by air drying. Cells were then stained with 0.5% (w/v) crystal violet in 20% methanol for 10 to 15 min with gentle rocking. Excess stain was removed by washing three to four times with distilled water. Plates were air-dried, and for quantification, the stain was solubilized with 10% acetic acid, incubated for 10 to 15 min, and absorbance was measured at 590.

EdU DNA Synthesis Monitoring Flow Cytometry.

K562 cells were pulse-labeled with 10 µM EdU for 20 min and SW480 and LS1034 for 40 min and subsequently fixed with 2% formaldehyde for 30 min at RT. EdU incorporation was detected using Click chemistry according to the manufacturer’s instructions (Click-iT EdU Flow Cytometry Cell Proliferation Assay, Invitrogen). Cells were resuspended in 1× PBS (BioConcept) with 1% bovine serum albumin (BSA), 2 µg/mL DAPI, and 0.5 mg/mL RNase A for 30 min at RT and subsequently analyzed on a BD LSR II (Becton Dickinson, USA) flow cytometer, using BD FACSDivaTM software, and quantified using FlowJo single-cell analysis software (FlowJo, LLC).

Western Blot.

Proteins were extracted from cells using a homemade RIPA buffer (50 mM Tris-HCl, 150 mM NaCl, 1% NP-40, 0.1% sodium dodecyl sulfate (SDS), and 0.5% sodium deoxycholate, pH 7.4) and quantified using the BCA Protein Assay Kit (Pierce). Equal amounts of protein were separated on a 4 to 20% gradient precast Tris-glycine gel from Thermo Fisher and transferred onto a nitrocellulose membrane using the iBlot 3 system from Thermo Fisher. The membrane was blocked with 5% nonfat milk in PBS containing 0.1% Tween-20 (PBS-T) for 1 h at RT, followed by overnight incubation at 4 °C with primary antibody as followed: Anti-HA-Peroxidase (High Affinity, 50 mU/mL, Roche, ref: 12013819001), horseradish peroxidase (HRP)-conjugated anti-actin antibody (1:5,000 dilution, ThermoFisher, ref: MA5-15739-HRP), Phospho-Chk1 (Ser345) (133D3) Rabbit mAb 2348, Phospho-Chk2 (Thr68) (C13C1) Rabbit mAb 2197, Phospho-ATM (Ser1981) (D6H9) Rabbit mAb #5883, Phospho-ATR (Ser428) Antibody #2853 and Phospho-Histone H2A.X (Ser139) (20E3) Rabbit mAb #9718, APOBEC3B (E9A2G) Rabbit mAb #41494, ORF1p (D3W9O) Rabbit mAb #88701. in 5% nonfat milk in PBS-T. After three washes with PBS-T, membranes were incubated with goat anti-rabbit antibody (1:5,000Anti-rabbit IgG, HRP-linked Antibody #7074). After three washes, protein bands were visualized using enhanced chemiluminescence detection reagent (Advansta Inc, ref: K-12049-D50) and imaged using a chemiluminescent imaging system (Fusion FX from Vilber).

Design of L1-ORF2p.

cDNAs for HA-tagged constructs were codon-optimized, synthesized using the GeneArt service from ThermoFisher, and cloned into the Dox-inducible expression vector pTRE-3HA with the sequences indicated (SI Appendix, Table S8).

OE Protocol for EN and EN^d in K562 Cells.

Dox-inducible OE was performed using a lentiviral (LV) system. K562 cells were transduced with LV particles containing EN-3XHA constructs under the control of a tetracycline-inducible promoter. Cells were seeded at 1 × 10⁵ cells/mL and transduced. After 48 h, cells were selected with 1 µg/mL puromycin for 3 d. OE was induced by adding 1 µg/mL Dox to the culture medium for 3 d. Induction efficiency was validated by western blot.

Cell Proliferation Assays for OE Experiments.

The PrestoBlue™ assay was used to assess the proliferation of K562 cells. Following 5 d of LV transduction and 3 d of puromycin selection, cells were either induced with Dox to express the transgene (+Dox) or left uninduced (−Dox). At each designated time point, 100 μL of cell culture was incubated with 10 μL of PrestoBlue™ reagent for 3 h at 37 °C. Absorbance was then measured immediately on a microplate spectrophotometer at 570 nm, with background absorbance at 600 nm subtracted for accuracy. Data normalization was conducted against the −Dox condition to account for baseline cell count variations.

CUT&Tag.

CUT&Tag was performed as described (58) without modifications. For each mark, 150 k cells were used per sample using the anti-H3K9me3 primary antibody (Active Motif, AB_2532132), the anti-H3K4me3 primary antibody (C42D8), and anti-rabbit IgG (Abcam, ab46540) secondary antibody. A homemade purified pA-Tn5 protein (3XFlag-pA-Tn5-Fl, Addgene #124601) was produced and coupled with Oligos by the Protein Production and Purification of EPFL, as previously described. Purified recombinant protein was used at a final concentration of 700 ng/μL (1:250 dilution from homemade stock). Libraries were sequenced with 75 bp paired end on the NextSeq 500 (Illumina). Reads were aligned to the hg19 reference genome using bowtie2 (59). Only proper read pairs with MAPQ > 10 were kept. Bigwig coverage tracks with the sum of replicate samples were generated using bedtools 2.27.168 (60) and deeptools 3.3.169 (56), and heatmap representations of the coverage signal were performed using computeMatrix function and plotHeatmap from deeptools 3.3.1. The following code was used.

RNA Extraction.

Total RNA extraction was performed using the NucleoSpin RNA plus kit (Macherey-Nagel) according to the manufacturer’s recommendations.

RNA-seq Libraries and Downstream Analyses.

RNA-seq libraries were prepared using the Illumina TruSeq Stranded mRNA kit. Libraries were sequenced in 75 or 100 bp paired-end formats on the Illumina HiSeq 4000 and NovaSeq 6000 sequencers, respectively. RNA-seq reads were mapped to the hg19 human genome releases using hisat v2.1.060. Only uniquely mapped reads were used for counting over genes and repetitive sequence integrants (MAPQ > 10). Counts for genes and TEs were generated using featureCounts v2 and normalized for sequencing depth using the TMM method implemented in the limma package of Bioconductor. Counts on genes were used as library size to correct both gene and TE expression. For repetitive DNA elements, an in-house curated version of the Repbase database was used. Differential gene expression analysis was performed using Voom61 as implemented in the Limma package of Bioconductor62. P-values were adjusted for multiple testing using the Benjamini–Hochberg method (SI Appendix, Table S6):

Genes were analyzed for ZNF93-HA binding proximity (9), and intersection between datasets was computed using the following website: https://bioinformatics.psb.ugent.be/webtools/Venn/. Gene analysis from the RNAseq is reported in SI Appendix, Table S7. Gene Set Enrichment Analysis for ZNF93 KD was done using the GSEA web interface (61, 62). The following options were used:

num, 100,

scoring_scheme, weighted,

norm, meandiv,

mode, Max_probe,

include_only_symbols, true,

set_max, 500,

gmx, ftp.broadinstitute.org://pub/gsea/gene_sets/h.all.v2023.2.Hs.symbols.gmt,

plot_top_x, 50,

nperm, 100,

order, descending,

rnd_seed, timestamp,

set_min, 15, res,

sort, real,

metric, Signal2Noise,

make_sets, true,

rnd_type, no_balance,

gui, false,

permute, phenotype,

collapse, No_Collapse

OE Protocol for ZNF93 in K562 Cells.

Dox-inducible OE was performed using a LV system. K562 cells were transduced with LV particles containing GFP-3XHA, LACZ-3XHA, or codon-optimized ZNF93-3XHA constructs under the control of a tetracycline-inducible promoter. Cells were seeded at 1 × 10⁵ cells/mL and transduced. After 48 h, cells were selected with 1 µg/mL puromycin for 3 d. OE was induced by adding 1 µg/mL Dox to the culture medium for 5 d. Induction efficiency was validated by western blot.

Generation of ZNF93^MUT.

ZNF93MUT-HA was generated by introducing four mutations into pTRE-ZNF93-HA via site-directed mutagenesis. The plasmid was digested with XhoI and NheI. Mutations were introduced using In-Fusion cloning with specific primers (SI Appendix, Table S7). PCR was performed with high-fidelity polymerase, followed by gel purification and recombination with the linearized plasmid using the In-Fusion HD Cloning Kit (Takara Bio©) according to the manufacturer’s protocol. The product was transformed into DH5α Escherichia coli cells, and colonies were screened by PCR. Positive clones were verified by Sanger sequencing, and confirmed ZNF93^MUT-HA plasmids were propagated and purified for further use.

Co-IP.

K562 cells overexpressing HA-tagged ZNF93, ZNF93^MUT, or GFP were harvested 72 h post–Dox induction, washed with ice-cold PBS, and lysed in 500 µL of immunoprecipitation whole cell lysate (IP-WCL) buffer (9.5 mL IP buffer base, 20 µL of 1 M benzamidine, 100 µL of 0.1 M methyl ethyl diketone sulfone (PMSF), 500 µL of Igepal 10% CA-630, and protease inhibitors). Lysates were sonicated (2 × 5 s, Amplitude 0.30) and clarified by centrifugation at 5,000 rpm for 5 min at 4 °C. The supernatant was collected for immunoprecipitation. HA-tagged proteins were immunoprecipitated by adding 25 µL of anti-HA magnetic beads (Pierce) per sample, which were resuspended in 800 µL of dynabuffer and washed once with 1 mL of dynabuffer and twice with 1 mL of IP-WCL buffer. Beads were incubated with 500 µL of lysate overnight at 4 °C with gentle rotation. After incubation, beads were washed three times with 800 µL of wash buffer (9.5 mL IP buffer base, 20 µL of 1 M benzamidine, 100 µL of 0.1 M PMSF, 500 µL of Igepal 10% CA-630, and protease inhibitors). For elution, beads were incubated with 62.5 µL of 1× loading buffer, heated at 95 °C for 10 min, and analyzed by SDS-polyacrylamide gel electrophoresis. Proteins were transferred to membranes, blocked with 5% milk or BSA in PBS-T, and probed with anti-KAP1 antibody (ab10483) overnight at 4 °C. After incubation with HRP-conjugated secondary antibody, signals were detected using chemiluminescence (Pierce). Input and were collected, treated with 12.5 µL of 5× loading buffer, heated, and analyzed in parallel to confirm the presence of KAP1 and HA-tagged proteins.

ChIP followed by qPCR (ChIP-qPCR).

K562 cells were cross-linked at RT for 10 min by adding formaldehyde to a final concentration of 1%, followed by quenching with glycine. Cells were washed twice with PBS, pelleted, and stored at −80 °C. Pellets were sequentially lysed in LB1 (50 mM HEPES-KOH pH 7.4, 140 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 10% glycerol, 0.5% NP-40, 0.25% Triton X-100, and protease inhibitors), LB2 (10 mM Tris-HCl pH 8.0, 200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, and protease inhibitors), and LB3 (10 mM Tris-HCl pH 8.0, 200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.1% NaDOC, 0.1% SDS, and protease inhibitors). Chromatin was sonicated (Covaris: 5% duty, 200 cycles, 140 PIP, 20 min) to obtain DNA fragments of 100 to 300 bp. ChIP was performed using an antibody against KAP1. Beads were coated with the antibody at 4 °C, followed by overnight incubation with chromatin at 4 °C. Immunoprecipitated complexes were washed sequentially with Low Salt Wash Buffer (10 mM Tris-HCl pH 8.0, 1 mM EDTA, 150 mM NaCl, and 0.15% SDS), High Salt Wash Buffer (10 mM Tris-HCl pH 8.0, 1 mM EDTA, 500 mM NaCl, and 0.15% SDS), LiCl Buffer (10 mM Tris-HCl pH 8.0, 1 mM EDTA, 0.5 mM EGTA, 250 mM LiCl, 1% NP-40, and 1% NaDOC), and TE buffer. DNA was eluted and purified using QIAGEN columns. ChIP and input DNA were analyzed by qPCR using primers reported (SI Appendix, Table S7). Enrichment was calculated as a percentage of input and is represented relative to control regions using the ΔΔCt method. At L1PA4 (SI Appendix, Fig. S4), the signal was normalized to the average signals of EVX1 (negative control region) and ZNF180 (positive control region). At the A3B promoter (Fig. 5), the average signal from four primers (ZNF93_p1 to ZNF93_p4) is shown alongside the average signal from four negative regions (EVX1 and Empty_p1 to Empty_p3) and one positive region (ZNF180).

Chemical Perturbation with HU.

Cells were cultured in the presence of HU (5 mM) for 24 h before western blotting analysis and release in fresh medium for subsequent FACS analysis.

Cell Cycle Analyses.

Cell cycle distribution was analyzed by flow cytometry measurement of cellular DNA content using PI staining. Three million cells were collected, washed, and resuspended in 1 volume of ice-cold 1× Dulbecco’s PBS (BioConcept) before being fixed by the addition of 2.5 volumes of ice-cold 100% ethanol during slow vortexing. After 45 min of incubation at 4 °C, cells were washed again in ice-cold PBS and resuspended in 1.5 volume of ice-cold PI staining solution (0.1% Triton, 200 μg/mL RNAse A, and 50 μg/mL PI). After 30 min of incubation at RT in the dark, the samples were analyzed on a Beckman Coulter Gallios flow cytometer, using Kaluza Analysis software, and quantified using FlowJo single-cell analysis software (FlowJo, LLC).

Data Analysis.

Unless otherwise specified, graphs were obtained by Excel software or ggplot2 R package.

Statistical Analysis.

All statistical tests and numbers of biological replicates are listed in the figure legends. All statistical tests were performed with R. Unless otherwise specified, stars represent P-value intervals: *< 0.05; **< 0.01; ***< 0.005.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (TXT)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Dataset S05 (TXT)

Dataset S06 (XLSX)

Dataset S07 (XLSX)

Dataset S08 (XLSX)

Data, Materials, and Software Availability

The accession number for the RNA-seq and CUT&Tag data generated in this article is GEO: GSE290774 (63). Publicly available datasets used in this study include TRIM28 ChIP-seq in H1 cells from P. Turelli, et al., (2014) accessible through the following number GSE57989 (64), KZFP ChIP datasets from Imbeault et al., (2017), Helleboid et al., (2019) and De Tribolet et al., (2023) available on KRABopedia https://krabopedia.org/, or through the following accession numbers GSE200964, GSE78099 and GSE120539 (65–67). Pan-cancer gene expression and proliferation correlation data from TCGA (https://portal.gdc.cancer.gov/) (68). Transcriptomic data from the TCGA CESC cohort were recovered from The HPV Induced Cancer Resource (THInCR) available at https://mymryklab.ca/main-home/ (69). RNAseq and ChIPseq in resting T-cell from Marzetta et al., (2019) are accessible through the following accession numbers GSE81871, GSE81872, and GSE81874 (70–72). All other data are included in the article and/or supporting information.