Transposable elements as novel therapeutic targets for PARPi-induced synthetic lethality in PcG-mutated blood cancer Open Access

Loss-of-function (LoF) mutations in Polycomb repressor complexes (PRCs) affecting enhancer of zeste homolog 2 (EZH2) and Additional sex combs like 1 (ASXL1) are frequently identified in both myeloid and lymphoid neoplasms, and correlate with poor prognosis.^1-6,ASXL1 exon 12 mutation resulting in frameshift and nonsense transcripts is one of the most common mutations accounting for up to 45% of patients with myelodysplastic syndrome (MDS)/myeloproliferative neoplasm (MPN) and chronic myelomonocytic leukemia (CMML).^1,7-9 Furthermore, ASXL1 LoF mutation has recently been reported as one of the most significant poor prognostic markers in patients with chronic lymphocytic leukemia (CLL).^10,11 Interestingly, forced expression of a truncated ASXL1 protein predicted from the frameshift mutation could also promote the development of myeloid malignancies in a transgenic mouse model, suggesting a potential pathological function of the putative truncated protein, although such proteins remain to be detected in patients.^12-15 Suppression of EZH2 activity has been proposed as a key oncogenic function associated with ASXL1 mutation,^15,16 which is consistent with the findings of EZH2 LoF mutation in ∼10% of patients with MDS/MPN and CMML.¹⁷ Paradoxically, in contrast to the classical mutual exclusivity of mutations with direct overlapping functions, EZH2 LoF mutation frequently coexists with ASXL1 mutation (cBioportal: MDS, adjusted P < .001; MPN, adjusted P = .032) in human patients with MDS/MPN^1,17 and CMML¹⁸ associated with a worse clinical outlook, urging the need for thorough experimental investigations to resolve this dilemma.^17,19,20 While it has been a major challenge to target LoF mutations by small molecule inhibitor approaches when mutated proteins often cannot be found and/or do not have a clear role in driving the disease, the lack of appropriate disease models has further compounded the issue and significantly impeded the progress in studying the underlying mechanisms of these diseases and development of novel effective therapeutics.

Transposable elements (TEs) that account for almost half of the human genome,^19,21-23 but are historically regarded as ancient junk sequences,^24,25 have been reported in recent years to be reactivated in driving disease development,²⁶ and multiple cellular processes including gene expression,²⁷ DNA damage, and immune responses.^28-33 To restrict and regulate their activities, TEs are normally suppressed by multiple epigenetic mechanisms including DNA methylation, histone modifications, and noncoding RNAs.²⁵ Although histone 3 lysine 9 (H3K9) methylation has been viewed as a major regulator for TE activity, recent literature also suggest a key regulatory function of H3K27 methylation by PRCs on TEs.^22,23 Consistently, TE expression can be altered as a result of epigenetic drug treatments in cancer.^26,34,35 Strikingly, TE reactivation in cancer cells can potentially drive anticancer immunity,^28-33 and generates novel immunogenic antigens,³⁴ although relevant preclinical and clinical therapeutic data are needed to assess their full potentials. On the other hand, reactivated TEs can also induce replicative stress and excessive DNA damage, which we hypothesize that may create a specific vulnerability and can potentially be exploited by synthetic lethality approaches.³⁶ 

The prevalence of LoF mutations affecting multiple Polycomb group (PcG) proteins in hematological malignancies provides an excellent opportunity to study and test our hypothesis. To this end, the current study modeled the loss of PcG mutations in hematopoietic stem/progenitor cells (HSPCs) that resulted in various hematological malignancies like those observed in humans. Importantly, we also demonstrate that specific epigenetic reactivation of TEs in these cells offers a novel therapeutic avenue for poly(ADP-ribose) polymerase inhibitor (PARPi)–induced synthetic lethality via suppression of PARP-mediated protection of single-strand DNA (ssDNA) break during target site–primed reverse transcription (TPRT), which can be abolished by reverse transcriptase inhibitors (RTis).

Frozen viable human samples were obtained from the UK CLL biobank and the King's College London (KCL) biobank with approval from the local ethics committee. All tissue samples would have gotten informed consent before banking and use.

Compound Asx1^fl/flEzh2^fl/fl Rosa26-CreER mice were generated by backcrossing Ezh2^tm1Tara (MGI2661097)³⁷ Gt(ROSA)26Sor^{tm1(cre/ESR1)Tyj} (MGI: 3699244) with Asxl1^{tm1c(EUCOMM)Wtsi38} in C57BL/6J (CD45.2) background, in which critical exons (exon 16-19 in Ezh2 and exon 3 in Asxl1) are flanked with loxP sites. All mice were held and kept in specific pathogen-free biological services unit at King’s College London, and experiments were performed according to the protocols in the project license (P8B15B1E1) under the Animals (Scientific Procedures) Act 1986.

Cell staining was performed as previously described³⁹ (supplemental Table 1, available on the Blood website).

The comet assay kit (catalog no. 4250-050-K; R&D Systems, Minneapolis, MN) was used according to the manufacturer instructions. When assessing the degree of DNA damage, at least 15 randomly selected microscopic fields were imaged per slide covering a minimum of 500 cells in total.

Staining was performed as previously described.⁴⁰ 

RNA sequencing (RNA-seq) and Cleavage Under Targets and Release Using Nuclease (CUT&RUN) were performed as previously described⁴¹ (supplemental Table 1). Next-generation sequencing (NGS) PE150 RNA-seq reads were aligned using the STAR aligner.⁴² TE locations were identified using Repeatmasker⁴³ and used by the featureCount of Rsubread⁴⁴ to give read counts at all known TEs. Differential expression was carried out using DESeq2.⁴⁵ CUT&RUN analysis was conducted using the CUT&RUNTools pipeline⁴⁶ with default parameters.

Detailed information is provided in supplemental Methods.

To study the impacts of loss of Asxl1 and Ezh2 on hematopoiesis, CD45.2⁺ bone marrow (BM) cells from single or compound Asxl1^fl/flEzh2^fl/fl mice harboring an inducible Rosa26-CreER allele were transplanted together with CD45.1⁺ control rescue cells into lethally irradiated CD45.1⁺ recipient syngeneic mice for assessment of their roles in normal and malignant hematopoiesis (Figure 1A). Although we observed myeloid-biased repopulation in the Ezh2 and Asxl1/Ezh2 knockout (KO) groups (supplemental Figure 1A-C), analysis of HSPCs uncovered that concurrent depletion of Asxl1 and Ezh2 enhanced the number of Lineage⁻c-kit⁺Sca1⁺ (LSK) cells, in particular multipotent progenitor 2 to multipotent progenitor 4 populations and myeloid progenitors (Lineage⁻c-kit⁺Sca1⁻) in BM (Figure 1B-C). Moreover Asxl1^−/−Ezh2^−/− LSK cells displayed accelerated cell cycle progression upon Ki67 staining (supplemental Figure 1D), partially accounting for the observed expansion of HSPCs.

Although the wild-type (WT) control animals did not develop any noticeable hematological malignancies, almost all Asxl1/Ezh2 double KO (dKO) mice succumbed to fatal MDS/MPN, associated with marked splenomegaly and extramedullary hematopoiesis in the presence of B-cell lymphoproliferative disorders (LPDs) in some animals (Figure 1D-H; supplemental Figure 1E-G; supplemental Table 2). This contrasts with the single KO mice in which Asxl1 KO mice were largely normal with mild polycythemia vera–like or essential thrombocythemia–like hematopoietic phenotypes noted in some animals, whereas Ezh2 KO resulted in low penetrant MDS/MPN LPDs with extended disease latency (Figure 1D-H, supplemental Figure 1E-G; supplemental Table 2). May-Grunwald Giemsa staining revealed markedly dysplastic features such as abnormal megakaryocytes (monolobated nuclei and bilobed megakaryocytes), giant platelets, Howell Jolly bodies, and pseudo–Pelger-Huët anomalies (hyposegmented neutrophils) in the BM and/or peripheral blood (PB) of Ezh2 single KO and Asxl1/Ezh2 dKO diseased mice, indicating trilineage dysplasia that was particularly noticeable in myeloid and megakaryocytic cell types (Figure 1E). Consistent with the respective disease phenotypes, Asxl1 KO diseased mice had elevated red blood cells, hemoglobin, and/or platelet levels in line with polycythemia vera–like or essential thrombocythemia–like phenotypes (Figure 1F; supplemental Figure 1E), whereas Asxl1/Ezh2 dKO showed left-shifted granulopoiesis with a significant increase in the percentage of myeloblasts and promyelocytes and a significant decrease of band cells (Figure 1G) with thrombocytosis (Figure 1F). Splenomegaly was clearly noted in mice that developed MDS/MPN with or without LPDs (Figure 1H). Together, these data indicate that inactivation of Asxl1 and Ezh2 in HSPCs results in development of aggressive myeloid malignancies as observed in human patients.

Further analysis of LPDs in the diseased mice revealed a major expansion of B220^−/low non-T(CD4⁻CD8⁻) lymphoid cells (Figure 2A; supplemental Figure 2A) that expressed (1) a specific gene signature matching those of B1α cells (supplemental Figure 2B), the putative origin of CLL^47,48; and (2) clonal immunoglobulin heavy variable domains (Figure 2B), suggesting clonal CLL-like disease. Indeed, we identified clonal rearrangement of IGHV-IGHD-IGHJ rearrangements (Table 1; supplemental Figure 2C-F) with unmutated IGHV genes (≥98% identity to germ line counterpart) in mice with CLL-like disease, indicating a monoclonal LPDs with unmutated IGHV genes. Intriguingly, 3 of 6 LPD mice shared an obvious stereotyped subset with highly homologous IGHV-IGHD-IGHJ gene rearrangement, displaying significant similarity on the variable domain heavy chain CDR3, a feature found in human patients with CLL,⁴⁹ suggesting a common antigen for the clonal expansion and pathological relevance of the PRC2-mutated CLL-like model, analogous to the corresponding human disease (Table 1).

To further assess their leukemic property, serial transplantation assay was performed using BM cells from primary hosts with WT, single KO, or dKO cells. Although no MDS/MPN-like disease phenotype was observed in mice that received secondary transplantation, aggressive CLL-like diseases developed with short latency in approximately half of the animals (6/13) transplanted with Asxl1^−/−Ezh2^−/− LPD cells but not Ezh2^−/− LPD cells (Figure 2C). The CLL mice with significant expansion of donor-derived B cells exhibited severe organomegaly (splenomegaly and hepatomegaly), high BM cellularity, leukocytosis, and anemia referring to a high-risk stage III according to Rai staging system⁵⁰ (Figure 2D; supplemental Figure 2G). Morphologically typical CLL cells with clumpy nuclei and a rim of visible cytoplasm were identified in the CLL mice (Figure 2E). Consistently, flow cytometry revealed the CLL with CD19⁺B220^−/lowCD5⁺IgM⁺CD43⁺ B1α cell phenotype (Figure 2F-G). Moreover, all CD19⁺ B cells were κ light chain restricted, indicating a clonal expansion of CLL cells (Figure 2H), which had the identical IGHV-IGHD-IGHJ gene rearrangement as the primary CLL cells (supplemental Figure 2H).

To further characterize the Asxl1^−/−Ezh2^−/− CLL model, we compared it with the commonly used Eμ-TCL1 transgenic CLL mouse model using BM cells for tertiary BM transplantation. Compared with mice with TCL1 cells that developed CLL at ∼12 weeks with moderate splenomegaly and a high level of CD19/CD5 double-positive cells in the PB, mice receiving either BM or splenocytes from Asxl1^−/−Ezh2^−/− CLL mice succumbed to disease in 6 weeks with organomegaly in the spleen, liver, and lymph nodes, and moderate level of circulating CD19/CD5 double-positive cells (Figure 2I; supplemental Figure 2I-K). In addition to lymphocytosis, anemia, and thrombocytopenia were also identified in Asxl1^−/−Ezh2^−/− CLL mice (supplemental Figure 2L). Together with the organomegaly, the aggressive phenotype of Asxl1^−/−Ezh2^−/− CLL could be defined as high-risk Rai stage IV in tertiary transplantation, whereas the tertiary transplanted TCL1 CLL phenotype was like that of patients with intermediate Rai stage II, without anemia and thrombocytopenia. Consistent with a more aggressive form of CLL,^51,52 cleaved CLL cells were only identified in Asxl1^−/−Ezh2^−/− CLL mice (supplemental Figure 2M). Together, these data depict a novel murine CLL model driven by loss of Asxl1 and Ezh2.

We performed RNA-seq analysis with BM (Figure 3A-C) or CD19⁺ splenocytes (supplemental Figure 3A-C) harvested from all 4 groups of mice that received primary transplant. Gene ontology and gene set enrichment analysis (GSEA) revealed that specific activation of apoptosis pathway found in single Asxl1^−/− and single Ezh2^−/− cells was absent in Asxl1^−/−Ezh2^−/− cells, which instead prominently expressed transcriptional programs with well-established roles in oncogenic transformation such as Myc and translation (Figure 3B-C; supplemental Figure 3B-C). Specific activation of these oncogenic pathways and suppression of apoptosis in Asxl1/Ezh2 dKO cell may explain the synergism between these 2 mutations. Interestingly, we also observed activation of multiple DNA damage repair (DDR) pathways in these cells, indicating excessive DNA damage, an established feature for reactivation of TEs²⁸ (Figure 3B-C; supplemental Figure 3B-C).

Gene set variant analysis was used as a complementary method to further assess their activities in BM CD19⁺ and c-kit⁺ HSPCs harvested from the BM 5 days or 3 months after tamoxifen treatment. As a result, Asxl1^−/−Ezh2^−/− cells consistently showed the highest expression of multiple DNA repair pathways in most of the comparisons except at the 5-day time point when transient Cre-recombination–mediated DDR was noted in all the floxed groups (Figure 3D-G). To further assess the specificity of the observed enrichment in DDR pathways, we compared RNA-seq from CD19⁺ spleen cells isolated from Asxl1/Ezh2 dKO or TCL1 CLL mice with WT CD19⁺ spleen cells. Although both CLL models shared several overlapping pathways (supplemental Figure 3D), DNA repair pathways were only found in Asxl1/Ezh2 dKO CLL (Figure 3H), which was further validated in GSEA analysis (Figure 3I), consistently suggesting excessive DNA damage being induced in Asxl1/Ezh2 dKO cells. Because DDR dysregulation can be potential therapeutic target for cancer,⁵³ we focused our analyses on the underlying mechanisms for deregulated DDR.

To this end, we closely examined candidate pathways previously reported being regulated by the PcG proteins including p53-p21–, coactivator-associated arginine methyltransferase 1-, and AKT serine/threonine kinase 1 mammalian target of rapamycin–dependent pathways, however, none of them was significantly deregulated in all relevant comparisons (supplemental Figure 3E, data not shown). Then we turned our attention to TEs with known capability of inducing DNA damage. We first mapped the RNA-seq data from BM cells (Figure 3J) or CD19⁺ splenocytes (Figure 3K) harvested from primary mice against all TEs in the genome to determine the differentially expressed families of TEs (adjusted P ≤ .01; log₂ fold change of ±1.5 or higher; maximum reads per million of ≥1000). Although a moderate upregulation on TE subfamilies were noted in single KO (Figure 3J), their expressions were significantly higher in Asxl1/Ezh2 dKO cells (Figure 3J-K; supplemental Figure 3F), suggesting that combined inactivation of both PcG genes leads to a strong reactivation of TEs. Whereas no DNA TEs was found to be upregulated, key family members of all 3 endogenous retrovirus (ERV) classes, including intracisternal A-type particles, MMERGLN, murine leukemia virus, and long-terminal repeat retrotransposon families were activated in Asxl1/Ezh2 dKO across different comparisons. We note that TE derepression was more significant in Ezh2 single mutant than those in Asxl1 single-mutant BM cells (Figure 3J-K). Importantly, the level of TE derepression correlated well with DDR, in which Ezh2 single-mutant BM cells showed stronger DDR activity than Asxl1 single mutant (Figure 3F), indicating a bigger contribution by EZH2 to TE depression and DDR in BM cells. However, the deregulation of TE expression by PcG is likely cell type specific because we observed a largely similar impact upon inactivation of individual PcG protein in CD19⁺ spleen cells (Figure 3K). Further analysis of the RNA-seq data from mice that received tertiary transplantation and developed advanced CLL confirmed that ERV expression was indeed specifically upregulated in Asxl1/Ezh2 dKO as compared with the control WT CD19⁺ B cells or TCL1 CD19⁺ CLL cells (Figure 3L). Mechanistically, CUT&RUN sequencing on CD19⁺ splenocytes revealed a decrease in H3K27 trimethylation repressive mark and an increase in the H3K27 acetylation activation mark for many ERV subfamilies differentially expressed in Asxl1/Ezh2 dKO (Figure 3M-N; supplemental Figure 3G), resulting in their transcriptional derepression. The epigenetic landscape of Asxl1/Ezh2 dKO CLL at TE elements is also different as compared with TCL1 CLL (supplemental Figure 3H), which had higher H3K27 trimethylation and lower H3K27 acetylation marks. These results consistently suggest epigenetic deregulation underlying the observed transcriptional derepression in Asxl1/Ezh2 dKO, which could be detected in the 3-month time point (supplemental Figure 3I-J), before notable disease onset. TPRT is a well-characterized mechanism for attempted reintegration of TEs into host genomes, which will produce ssDNA intermediates that are protected by PARP and replication protein A complex.⁵⁴ Given that PARP activity is required to safeguard transiently exposed ssDNA during TPRT, we hypothesize that suppression of PARP may induce excessive DNA damage in these cells and represent an effective novel treatment for the diseases.

In order to test this hypothesis, we first assessed the nature of DNA breaks in dKO cells by neutral and alkaline comet assays. Consistent with our hypothesis, we observed that although similar level of double-strand DNA (dsDNA) breaks found in both Asxl1/Ezh2 dKO and WT control, Asxl1/Ezh2 dKO cells however carried a significantly higher level of ssDNA breaks during TPRT, which are protected by PARP from progressing into devastating dsDNA breaks (supplemental Figure 4A). Then we investigated whether Asxl1^−/−Ezh2^−/− c-kit⁺ HSPCs (Figure 4A-D) were indeed sensitive to PARPi (olaparib) treatment using in vitro culture and clonogenic assay as previously described.⁴⁰ Although a concentration of up to 1μM olaparib exhibited minimal effects on normal HSPCs (supplemental Figure 4B), Asxl1^−/−Ezh2^−/− cells were highly sensitive to PARPi-induced DNA damage as evidenced by detection of excessive DNA damage gH2Ax foci (Figure 4A) and dsDNA breaks as showed in comet assays (Figure 4B-C; supplemental Figure 4C). Importantly, PARPi resulted in ∼90% reduction in their colony numbers (Figure 4D) but had little or no detectable effect on WT or control PARPi-resistant MLL-fusion transformed cells⁴⁰ (Figure 4A-D; supplemental Figure 4C-D). Similar results were obtained using a different PARPi, veliparib that also exhibited specific suppression on Asxl1^−/−Ezh2^−/− but not WT HSPCs (supplemental Figure 4E). Next, we tested the PARPi vulnerability of Asxl1^−/−Ezh2^−/− cells using another new Asxl1^−/−Ezh2^−/− Runx1KD mouse model that mimic RUNX1 haplodeficiency frequently coexisting with the PcG LoF mutations in patients with MDS.¹ Knocking down the expression of Runx1 in Asxl1^−/−Ezh2^−/− c-kit⁺ cells produced an even more aggressive MDS/MPN disease with a shorter latency than Asxl1^−/−Ezh2^−/− control (supplemental Figure 4F-H). Strikingly, PARPi treatment very significantly suppressed the colony-formation ability in Asxl1^−/−Ezh2^−/−Runx1KD MDS cells and again had no notable impact on control MLL-fusion transformed cells (Figure 4E), suggesting a specific PARPi vulnerability was created and remained in Asxl1^−/−Ezh2^−/− cells despite the presence of additional genetic events.

In addition to myeloid leukemia, we also tested the PARPi sensitivity of CLL cells (Figure 4F-L). Although olaparib treatment did not exert any notable effect on TCL1 CLL cells, it significantly induced apoptosis and reduced the number of Asxl1^−/−Ezh2^−/− CLL cells (Figure 4F; supplemental Figure 4I). Similar results were observed using 2 independent PARPis, veliparib and talazoparib (supplemental Figure 4J). Consistently, significant enrichments of γH2AX DNA damage foci (Figure 4G) and dsDNA breaks (Figure 4H-I) were specifically observed in Asxl1^−/−Ezh2^−/− cells as compared with untreated controls and TCL1 CLL cells upon PARPi treatment (Figure 4H-I; supplemental Figure 4K). To further elucidate the in vivo efficacy of PARPi treatment, mice transplanted with Asxl1^−/−Ezh2^−/− CLL were treated with olaparib 1 week after tertiary transplantation (Figure 4J). Blood analyses at day 35 (end of treatment) revealed that although red blood cell, hemoglobin, and platelet counts in olaparib treatment groups were similar to untreated controls, the CLL counts in PARPi-treated animals were significantly reduced (supplemental Figure 4L). Furthermore, in week 4, mice treated with vehicle started developing severe organomegaly and circulation of CLL cells in the PB, whereas olaparib significantly delayed the disease onset (Figure 4J-L). Together, these results reveal a specific DNA damage vulnerability in Asxl1^−/−Ezh2^−/− leukemia cells that can be targeted by PARPi.

Mechanistically, PARPi sensitivity can be due to suppression of homologous recombination (HR) genes^55,56 as we have previously reported for acute myeloid leukemia (AML) driven by repressive oncogenic transcription factors such as RUNX1/RUNX1T1 (AML1-ETO).⁴⁰ However, in line with our reactome, GSEA and differential gene expression analyses in which we found that key HR genes including Brca1 and Rad51 were mostly unchanged or even upregulated in dKO cells (supplemental Figure 3E), functional studies assessing Rad51 repair foci (supplemental Figure 5A), parylation (supplemental Figure 5B), and HR functionality (supplemental Figure 5C) further demonstrated HR competency in Asxl1^−/−Ezh2^−/− dKO cells, consistently indicating an alternative underlying mechanism, which we hypothesize is caused by reactivation of TEs. Reverse transcriptase is an essential enzyme specifically required for the replication life cycle of retroviral TEs and TPRT.^24,25 To confirm the functional link between TE reactivation and PARPi sensitivity, we first sought to investigate whether inhibition of the key reverse transcription step by 2 independent nucleoside/nucleotide RTis, namely didanosine and lamivudine, would suppress the observed PARPi sensitivity. As expected, c-kit⁺–enriched HSPCs from Asxl1^−/−Ezh2^−/− mice but not WT mice were highly sensitive to PARPi treatment (Figure 5A-C; supplemental Figure 5D-F). Strikingly, PARPi sensitivity was partially but significantly reversed by cotreatment with either dianosine or lamivudine, which otherwise had no impact on HSPCs derived from WT mice. Importantly, RTi treatment did not affect PARPi sensitivity of AML1-ETO–transformed cells (supplemental Figure 5G-I), whose PARPi sensitivity results from suppression of HR genes,⁴⁰ consistently indicating that RTis do not generally counteract PARPi treatment but exert a specific function targeting TPRT in Asxl1^−/−Ezh2^−/− dKO cells. Conversely, excessive γH2AX DNA damage foci and dsDNA breaks induced by PARPi in Asxl1^−/−Ezh2^−/− cells were partially but significantly reversed by RTis, suggesting that reactivation of TE plays a key role in the observed PARPi sensitivity (Figure 5A-C). Importantly, Asxl1^−/−Ezh2^−/− Runx1KD cells remain PARPi sensitive in a RTi-dependent manner (supplemental Figure 5J), suggesting that additional mutations do not diminish the key role of reactivated TEs in the observed PARPi sensitivity. Next, we performed similar treatments on Asxl1^−/−Ezh2^−/− CLL cells (Figure 5D-G). As expected, PARPi treatment was effective in suppressing the CLL cells by inducing excessive DNA damage (Figure 5D-F; supplemental Figure 5K). Similarly, we observed marked rescue of PARPi killing in Asxl1^−/−Ezh2^−/− CLL cells by different RTi, which resulted in reduced DNA damage and improved CLL cells survival upon PARPi treatment (Figure 5D-F). Finally, we also assessed in vivo combination treatment on mice transplanted with Asxl1^−/−Ezh2^−/− CLL cells. As a result, although RTi alone did not affect the disease latency consistent with a limited role of TE reactivation in driving disease development, it dramatically reversed efficacy of PARPi treatment and shortened the survival time, which was virtually identical to the PARPi-untreated control group (Figure 5G; supplemental Figure 5L). Together, these results demonstrate a direct functional link between reactivation of TEs and PARPi sensitivity in Asxl1^−/−Ezh2^−/− leukemia cells.

To examine whether similar findings in our mouse models can also be translated to human leukemia, we examined the TE expression profile of samples from patients with MDS carrying ASXL1/EZH2 mutations.⁵⁷ In line with our mouse models, we observed (1) preferential upregulation of retroviral TEs, LINE1 (Figure 6A; supplemental Figure 6A-D), which is the only major active TEs that can elicit DDR in the human genome^24,25; and (2) activation of DDR pathways in patient samples with Asxl1/Ezh2 mutations (Figure 6B). Consistently, at the functional level, these ASXL1/EZH2-mutated MDS samples were also sensitive to PARPi treatment (Figure 6C), which induced excessive γH2AX DNA damage foci (Figure 6D) and dsDNA breaks (Figure 6E; supplemental Figure 6E) in the PcG mutant but not in control MLL-rearranged samples. Comparable inhibitory effects were also observed using different PARPis, including veliparib and talazoparib (supplemental Figure 6F-I). Next, we performed similar analyses on samples from human patients with CLL with ASXL1 mutation compared with samples from those with non–ASXL1-mutated CLL reported in previous studies by us and others.^10,11 In line with our findings in MDS samples, we observed activation of LINE1 (Figure 6F) and DDR pathways (Figure 6G) in ASXL1-mutated CLL. Importantly, cells from patients with ASXL1-mutated CLL were also sensitive to PARPi treatment that was otherwise ineffective on non–ASXL1 (TP53)-mutated CLL cells (Figure 6H). Moreover, PARPi treatment induced significant DNA damage in ASXL1-mutated CLL but not WT CLL (Figure 6I; supplemental Figure 6J). Consistently, the observed PARPi sensitivity as in our mouse models could also be partially reversed by RTis (Figure 6J-K; supplemental Figure 6K). Together with the mouse data, these findings consistently indicate that reactivation of TEs as a result of LoF mutation in PcG proteins provides a novel vulnerability for PARPi-induced synthetic lethal targeting of these cancers.

In contrast to gain-of-function mutations, LoF mutations are typically intractable by classical small molecule inhibitor approaches. Synthetic lethality can be an ideal approach to target mutations leading to compromised DNA repair responses such as LoF BRCA mutations in ovarian and breast cancers.^58,59 However, mutations affecting DDR genes are not common in many cancers including leukemias, which are frequently driven by mutations affecting transcriptional and epigenetic gene regulations.^40,56,60 We and others have previously shown that oncogenic drivers can confer PARPi sensitivity to leukemia cells because of their impacts on HR (eg, AML1-ETO, promyelocytic leukemia-retinoic acid receptor alpha, ten-eleven translocation 2),^40,61 or can be targeted when PARPis are used in combination with other therapeutics (eg, MLL fusions).^40,62,63 In this study, we reveal a novel application of PARPi-induced synthetic lethality by targeting TPRT associated with reactivated TEs in leukemia driven by PcG LoF mutations (Figure 7), which is associated with very poor prognosis and do not have effective therapies. Although the ubiquitous and abundant nature of TEs present in cancer cells can have great potentials as molecular targets, very little progress has been made in deriving strategies considering TE reactivation for cancer therapeutics.⁶⁴ To date, the only and most well-documented potential avenue is based on putative immunological responses triggered by reactivated TEs^28-34,65,66; however, many patients with cancer still fail to mount proper immune responses because of cell intrinsic and/or extrinsic factors such as tumor microenvironment.^67-69 

After genetic losses of major epigenetic regulators commonly found in hematological malignancies, we showed that inactivation of PcG resulted in epigenetic reactivation of TEs and DNA damage in both mouse models and primary human patient samples. Strikingly, although reactivated TEs do not seem required for leukemic transformation, they create a vulnerability to PARPis. Consistently, PARPi treatment alone can significantly suppress leukemia cell growth and extend the disease latency. PARPi-induced DNA damage and selective killing can be partially blunted by RTis that suppress the formation of complementary DNA intermediate, key for TE life cycle PRC-mutated cells (Figure 7). The mechanism underlying the observed synthetic lethality by targeting specific functions of PARP in TPRT is intriguing and distinct from the classical PARPi sensitivity mediated by HR deficiency.^40,70 Although future mechanistic and orthogonal studies (eg, CRISPR interference targeting specific classes of TE subfamilies, and integrase inhibitor studies) are needed to further pinpoint the subfamilies of TEs and dissect key molecular players mediating the PARPi phenotypes, this study sets the stage for a novel and broader approach of creating synthetic lethality for human cancers (Figure 7).

Evidently, the expression of TEs is tightly regulated by epigenetic (eg, PIWI-interacting RNA, DNA methylation, and histone modifications) and transcription (eg, KRAB-containing zinc-finger proteins) factors.²⁵ Although spatial- and temporal-specific gene regulations are likely governed by a combination of transcriptional and epigenetic regulators, it is tempting to speculate potential utility of PARPis in cancers carrying epigenetic mutations (eg, DNMT, SETDB1, and PRC) that may lead to similar derepression of TEs. Future studies examining the different cell type–specific mechanisms of epigenetic regulations of TEs will shed light on this issue and help to further broaden the application of PARPi-induced synthetic lethality in other cancers.

The authors thank Kostas Stamatopoulos for stereotype analysis and comments on the manuscript; Alexander Tarakhovsky for provision of Ezh2 fl/fl mouse; members of the laboratory of Chi Wai Eric So for critical discussion; Jon Strefford for advice on chronic lymphocytic leukemia (CLL) sample analysis; and KCL Biobank, UK CLL Biobank, and the German Studienallianz Leukämie Akute Myeloische Leukämie Biobank for patient samples. The authors also thank Asmaa Elshahawy, Amelia Rushton, Rajani Chelliah, Austin Kulasekararaj, Melanie Oates, Pauline Robbe, Heidi Altmann, and Martin Bornhäuser for sample collection and annotation; and Biomedical Research Centre and Computational Research, Engineering and Technology Environment at KCL for technical support.

C.W.E.S. is a Royal Society Wolfson Research Merit Award holder. This work was supported by Cancer Research UK program grant (C17197/A29213); National Institutes of Health, National Cancer Institute R01 grant (R01 CA264932), Schweizerische Nationalfonds Sinergia grant (CRSII5_213586/1), Research Grant Council Theme–based Research Scheme (T12-702/20-N), and Blood Cancer UK grant (20002).

Contribution: B.B.Z. and C.-T.T. performed and executed most of the experiments and analyzed the data; T.K.F., C.V., E.T., B.J., P.N.I.L., N.L., A.C., L.B., A.T.A., A.G., J.L., S.-Y.Z., and C.L.F. helped with execution of experiments and data analyses; I.R.T. and C.L. helped with bioinformatic analyses; R.H., P.E.M.P., and A.S. advised on chronic lymphocytic leukemia biology and data interpretation; G.M. advised on the clinical features and interpretation of myelodysplastic syndrome (MDS) models; S.O. provided and assisted human MDS data set analyses; M.M.K. advised on transposable element analysis and interpretation; and C.W.E.S. was responsible for the overall design of the project and preparation of the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Chi Wai Eric So, Comprehensive Cancer Centre, Kings College London, 123 Coldharbour Ln, London SE5 9NU, United Kingdom; email: eric.so@kcl.ac.uk.